Audio Encoders, Audio Decoders, Methods and Computer Programs Adapting an Encoding and Decoding of Least Significant Bits

Technical Field

Embodiments according to the invention are related to audio decoders for providing a de-coded audio information on the basis of an encoded audio information.

Further embodiments according to the invention are related to audio encoders for provid-ing an encoded audio information on the basis of an input audio information.

Further embodiments according to the invention are related to methods for providing a decoded audio information on the basis of an encoded audio information.

Further embodiments according to the invention are related to methods for providing an encoded audio information on the basis of an input audio information.

Further embodiments according to the invention are related to respective computer pro-grams.

Embodiments according to the invention are related to an improved truncation of arithme-tic encoded audio data.

Background of the Invention

In the past, many different concepts for the encoding and decoding of audio content have been developed.

For example, the New Bluetooth Codec (NBC) is an audio codec which is very similar to a MDCT-based TCX audio codec used in the 3GPP EVS standard [1]. Both employ scalar quantization and context-based arithmetic encoding (confer, for example, references [2] to [4]) for coding the MDCT data.

The scalar quantizer is a simple uniform quantizer (with an additional dead-zone) whose step size is controlled by a unique global-gain (which is, for example, sent to the decoder as side information). This global gain controls both the distortion introduced by the scalar quantizer and also the number of bits consumed by the arithmetic encoder. The higher the global-gain is, the higher is the distortion and the lower is the number of bits consumed by the arithmetic encoder.

In EVS, like in most other communication codecs, the codec bitrate is constant, i.e., there is a limited number of bits (bit budget) available for encoding the MDCT data.

Consequently, the encoder should find (or has to find) a global-gain which is not too low, otherwise the number of bits consumed by the arithmetic encoder would exceed the bit budget. Also, it should (or has to) find a global-gain which is not too high, otherwise the distortion introduced by the quantization would be higher, resulting in worse perceptual quality of the decoded output signal.

Ideally, the encoder should find at every frame the optimal global-gain: the one which gives minimum distortions while producing a number of bits below the bit budget.

This goal can, for example, be achieved using an iterative approach known also as rate-loop: at every iteration of the loop, the MDCT data is re-quantized, the number of bits con-sumed by the arithmetic encoder is estimated and the global gain is adjusted as a function of the number of bits and/or the distortion.

A rate-loop is, however, computationally complex, and to save complexity, usually a small number of iterations is used. This is particularly relevant for very low power communica-tion codecs (for example, the New Bluetooth Codec) which require very low computational complexity. So, in practice, a suboptimal global-gain is usually found.

It has been found that in some cases, the found global-gain is too high, resulting in a con-sumed number of bits significantly lower than the bit budget. In this case, there is a num-ber of unused bits. These bits can actually be used by an additional tool called "residual quantization/coding" (which is, for example, used in EVS and NBC). This tool refines the quantized non-zero coefficients using one bit pro coefficient, and helps getting a distortion which is not too high even when the global-gain is too high.

Moreover, it has been found that, in some other cases, the found global-gain is too low, resulting in a consumed number of bits exceeding the bit budget. In this case, the quan-tized data cannot be entirely encoded. In other words, a portion of the data has to be left out in order to stay within the bit budget. A solution employed in the EVS standard (and also currently in NBC) is to truncate the high-frequency non-zero coefficients, by setting them to zero. Since the arithmetic encoder does not encode the portion of high-frequency zero coefficients (by using a last-non-zero-coefficient index), this approach allows saving bits and if enough high-frequency non-zero coefficients are truncated, this allows to stay within the bit budget.

It has been found that this approach is producing good results at low bitrates because the high-frequency coefficients are perceptually less important and they can be replaced by a random noise (using a noise filling tool, see for example, EVS [1]) without a significant loss in perceptual quality.

However, it has also been found that, at high bitrates, this approach can severely degrade the codec performance.

In view of this situation, there is a desire to have a concept which allows for an improved tradeoff between audio quality, complexity and bitrate.

Summary of the Invention

An embodiment according to the invention creates an audio decoder for providing a de-coded audio information on the basis of an encoded audio information. The audio decoder is configured to obtain the decoded spectral values on the basis of an encoded infor-mation representing these spectral values. The audio decoder is configured to jointly de-code two or more most significant bits per spectral value (for example, per quantized spectral value) on the basis of respective symbol codes for a set of spectral values using an arithmetic decoding. A respective symbol code represents two or more most significant bits per spectral value for one or more spectral values. The audio decoder is configured to decode one or more least significant bits associated with one or more of the spectral val-ues in dependence on how much least significant bit information is available, such that one or more least significant bits associated with one or more of the spectral values (which may, for example, be quantized spectral values) are decoded, while no least signif-icant bits are decoded for one or more other spectral values for which two or more most significant bits have been decoded and which comprise more bits than the two or more most significant bits. Moreover, the audio decoder is configured to provide the decoded audio information using the (decoded) spectral values.

This audio decoder allows for an efficient encoding/decoding concept which provides for a good tradeoff between audio quality, complexity and bitrate. For example, the audio de-coder can well-handle cases in which a bit budget is insufficient in order to encode all (quantized) spectral values at the side of an audio encoder under a given bit budget con-straint.

The audio decoder is based on the finding that, for a given bit budget, a comparatively good audio quality can be achieved if one or more most significant bits are encoded (and decoded) for many spectral values (or even for all non-zero spectral values) while omitting the encoding (and the decoding) of the least significant bits of some of the (quantized) spectral values. In other words, it is the key idea of the present invention that a degrada-tion of an audio quality in a case in which a bit budget is insufficient (for example, for a full encoding of quantized spectral values) is often smaller if the encoding and decoding of some least significant bits is omitted when compared to a solution in which an encoding of full spectral values is omitted. Worded differently, it has been found that omitting an en-coding of least significant bits of many spectral values is typically still a better solution to reduce a bit demand (to keep within a bit budget) when compared to completely omitting an encoding of a comparatively smaller number of spectral values (even if only spectral values in a high frequency region would be omitted). Wording it differently, the present invention is based on the finding that (selectively) omitting a decoding of least significant bits for spectral values, for which the one or more most significant bits have been decoded is a good way to reduce a bit demand which typically brings along less distortions when compared to an omission of the encoding and decoding of spectral values in a high fre-quency range.

Accordingly, the audio decoder described here typically does not bring along severe sig-nal-to-noise-ratio degradations in frames in which a bit budget is insufficient for a full loss-less encoding of quantized spectral values.

Moreover, it has been found that the concept is particularly efficient in a case in which two or more most significant bits per spectral value are jointly encoded and decoded, because in this case the most significant bits carry a sufficiently meaningful information in order to allow for a good audio representation even in the case that the least significant bits are not encoded and decoded. In other words, by jointly decoding two or more most signifi-cant bits per spectral value, it can be ensured that there are no excessive artifacts, which would be caused, for example, by introducing audio content encoded with less than two bits in a high frequency region. In other words, it has been found that the concept men-tioned herein provides for a good comprise between bitrate, complexity and audio quality.

In a preferred embodiment, the audio decoder is configured to map one symbol of an arithmetically encoded representation, which represents at least two most significant bits of at least one spectral value, onto the at least two most significant bits of the at least one spectral value. Accordingly, it can be achieved that the two or more most significant bits are represented by a single symbol of the arithmetically encoded representation (which is part of the encoded audio information), which allows for a good consideration of an encod-ing/decoding context and of statistical dependencies between adjacent (quantized) spec-tral values.

In a preferred embodiment, the arithmetic decoding is configured to determine bit posi-tions (for example, bit weights) of the at least two most significant bits (for example, des-ignated herein as "numbits" and "numbits-1") and to allocate the at least two most signifi-cant bits determined by a symbol of the arithmetically encoded representation to the de-termined bit positions. The bit positions can be determined, for example, on the basis of a number of so-called "escape symbols", which may also be designated as "VAL_ESC". For example, the bit positions may be determined individually for different symbols of the arithmetically encoded representation. Accordingly, a proper numeric weight can be allo-cated to the most significant bits, and it can also be found out as to whether one or more least significant bits and one or more intermediate bits (bit positions of which are between the one or more least significant bits and the two or more most significant bits) are associ-ated with a spectral value. Thus, it can be decided whether there should still be a decod-ing of one or more least significant bits for the respective spectral values (and, optionally, of one or more intermediate bits for the respective spectral value). Also, by using this con-cept, it is possible to avoid an encoding and decoding of least significant bits for such spectral values for which the two or more most significant bits are sufficient to fully repre-sent the spectral value. This is, for example, true for spectral values lying within a range between 0 and 3 (in the case that there are two most significant bits).

In a preferred embodiment, the audio decoder is configured to decode, for all spectral values for which two or more most significant bits have been decoded and which comprise more bits than the two or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the two or more most significant bits. Accordingly, it is possible to decode all bits of a binary number representation of a quantized spectral value, except for the least significant bit. For exam-ple, it is possible to decode all bits of the binary (and possibly signed) number representa-tions of all spectral values, with the exception of the least significant bit, for all non-zero spectral values. Thus, a good representation of the spectrum can be obtained, wherein it is ensured that a maximum error for each spectral value is limited to the least significant bit, independent from the question whether the encoded representation of the least signifi-cant bit for the respective spectral value can be included into the encoded audio represen-tation due to the bitrate constraints or not.

In a preferred embodiment, the audio decoder is configured to decode, in a first decoding phase (for example, step 3 of the decoding), two or more most significant bits per spectral values, and for all spectral values for which two or more most significant bits are decoded and which comprise more bits than the two or more most significant bits (which are jointly decoded) and a least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the two or more most significant bits. Moreover, in the first decoding phase, for all spectral values for which two or more most significant bits are decoded and for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, signs are decoded. Moreover, the audio decoder is configured to selectively omit, in the first decoding phase, a decoding of a sign for spectral values for which the two or more most significant bits, and any intermediate bits, as far as intermediate bits are present, indicate a zero value. Moreover, the audio decoder is configured to selectively obtain, in a second decoding phase (for example, step 6 of the decoding) which follows the first decoding phase, sign information for spectral values for which the two or more most significant bits and any in-termediate bits - as far as intermediate bits are present - indicate a zero value and for which a least significant bit information indicates a non-zero value.

Accordingly, no sign decoding is performed in the first phase if those bits decoded in the first phase (namely the two or more most significant bits and any intermediate bits which may be present) indicate that an absolute value of the spectral value is not larger than a contribution of a least significant bit. Thus, the decoding of the sign is postponed until the actual decoding of the least significant bit. Such a procedure is advantageous, since it can be avoided that a sign is decoded "too early" and in vain, which could be the case if the least significant bit corresponding to the respective spectral value is not included in the bitstream due to an exhaustion of a bit budget.

In a preferred embodiment, the audio decoder is configured to sequentially use subse-quent bits of a least-significant-bit-information bit sequence (for example, Isbs[]) in order to obtain least significant bit values associated with the spectral values. Accordingly, it can be achieved that there is a contiguous bit sequence which represents the least significant bits (any signs, as far as necessary). By shortening this bit sequence (e.g. Isbs[]), a re-quired bitrate for the transmission of the encoded audio representation can easily be ad-justed at the side of an audio encoder, and the audio decoder can very easily, and without a complex bit mapping, adapt to such an adjustment of the bitrate (or to a variable length or Isbs[]).

In a preferred embodiment, the audio decoder is configured to use a single bit (e.g. step 6, bit0) of the least-significant-bit-information bit sequence (e.g. Isbs[]) for respective spectral values for which the two or more most significant bit values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the single bit of the least-significant-bit-information bit sequence is used in order to obtain a least significant bit value in this case. Moreover, the audio decoder is configured to use a single bit (e.g. step6, bit0) of the least-significant-bit-information bit sequence for respective spectral val-ues for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which the used single bit of the least-significant-bit-information bit sequence confirms the zero value (e.g. value "0" of bit 0 in step 6). Moreover, the audio decoder is configured to use two subsequent bits (e.g. bit 0 and bit 1 in step 6) of the least-significant-bit-information bit sequence for re-spective spectral values for which the two or more most significant values and any inter-mediate bits, as far as intermediate bits are present, indicate a zero value, and for which a first of the used bits of the least-significant-bit-information bit sequence indicates a devia-tion from the zero value by a least significant bit value (value "1" of bit0 in step 6), where-in a second of the used bits (e.g. bit 1 in step 6) of the least-significant-bit-information bit sequence determines a sign of the respective spectral value.

By using such a mechanism, a high bitrate efficiency can be achieved. There is only one contiguous bit sequence (e.g. Isbs[]) for the encoding and decoding of the least significant bits, wherein this one contiguous bit sequence also selectively contains sign information for such spectral values which only deviate from a zero value by a least significant bit val-ue (i.e., for which the two or more most significant bits and any intermediate bits (as far as intermediate bits are present) indicate a zero value).

In a preferred embodiment, the audio decoder is configured to decode least significant bits starting from a least significant bit associated with a lowest frequency spectral value and proceeding towards spectral values associated with increasingly higher frequencies, such that spectral values (for example, all spectral values which comprise more bits than the two or more most significant bits) are refined by a least significant bit information in a range from a lowest frequency spectral value up to a spectral value for which a last least significant bit information is available, and such that (for example, all) spectral values (for example, even decoded spectral values which comprise more bits than the two or more most significant bits) having associated frequencies higher than a frequency associated with the spectral value for which the last least significant bit information is available remain unrefined. In other words, spectral values in a lower frequency range (from the lowest frequency spectral value up to a spectral value having associated the last least significant bit information) are refined using a least significant bit information, while spectral values associated with higher frequencies all remain unrefined. Consequently, a resolution in the perceptually more important low frequency range is increased by using a least significant bit refinement, while only the two or more most significant bits (and intermediate bits, if available) are used in a higher frequency range, which is perceptually less important. Consequently, the best possible hearing impression can be obtained on the basis of the available bitrate, wherein there is also a simple mechanism for which spectral values least significant bit information is provided. Furthermore, the spectral values can be refined from a lowest frequency spectral value up to a spectral value to which the last least signif-icant bit information is associated.

In a preferred embodiment, the audio decoder is configured to be switchable between a first mode, in which a decoding of spectral values in a higher frequency range is omitted (for example, entirely omitted) in response to a signaling from the encoder and in which least significant bits are decoded for all spectral values for which one or more most signifi-cant bits are decoded and which comprise more bits than the most significant bits, and a second mode, in which one or more least significant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits are decoded and which comprise more bits than the most significant bits.

In other words, the audio decoder is switchable between two modes, which use signifi-cantly different mechanisms for handling an exhaustion of a bit budget.

In the first mode, all spectral values in a lower frequency range are encoded (and decod-ed) fully including the least significant bit, while all spectral values in a higher frequency range are entirely discarded by the encoder even if they are non-zero and consequently not decoded at the side of the decoder. In the second mode, at least the most significant bits are encoded for all non-zero spectral values (and thus also decoded), but the least significant bits are only encoded (and decoded) if (or as long as) there is still a bit budget available.

However, it has been found that the possibility to switch between the two different modes allows the audio decoder to adapt to varying transmission conditions. For example, it has been found that the first mode is sometimes more advantageous than the second mode, for example if there is only a very small bitrate available. On the other hand, it has also been found that the first mode does not provide for good results in the presence of a suffi-ciently high bitrate, where the binary representations of many spectral values comprise least significant bits in addition to the two or more most significant bits. Accordingly, the audio decoder can operate with good results under circumstances in which there are only a few least significant bits and under circumstances where there is a comparatively large number of least significant bits (wherein the operation in the second mode is typically problematic in the first case, while the operation in the second mode is typically very ad-vantageous in the second case).

In a preferred embodiment, the audio decoder is configured to evaluate a bitstream flag which is included in the encoded audio representation in order to decide whether the au-dio encoder operates in the first mode or in the second mode. Accordingly, the switching between the first mode and the second can be controlled by an audio encoder, which typi-cally comprises good knowledge about which mode is most advantageous. Also, the complexity of the audio decoder can be reduced, because the audio decoder does not need to decide by itself whether to use the first mode of the second mode.

In another embodiment, the audio decoder is configured to jointly decode two or more most significant bits per spectral value for at least two spectral values on the basis of re-spective symbol codes, wherein a respective symbol code represents two or more most significant bits per spectral value for at least two spectral values. Such a grouping of spec-tral values, wherein two or more spectral values are represented by a single symbol of the arithmetically encoded representation is also particularly efficient, because there is often some correlation between adjacent spectral values, and because it is not necessary to

individually encode the bit position for each of two most significant bits. However, it can naturally happen that the "most significant bits" of one of the spectral values are both "ze-ro", because the bit position is typically determined by the spectral value having a larger absolute value.

An embodiment according to the invention creates an audio decoder for providing a de-coded audio information on the basis of an encoded audio information. The audio decoder is configured to obtain decoded spectral values on the basis of an encoded information representing the spectral values. The audio decoder is configured to decode one or more most significant bits on the basis of respective symbol codes for a plurality of spectral val-ues, and to decode one or more least significant bits for one or more of the spectral val-ues. In particular, the audio decoder is configured to be switchable between a first mode, in which a decoding of spectral values in a higher frequency range is omitted (for exam-ple, entirely omitted) in response to a signaling from the encoder and in which least signif-icant bits are decoded for all spectral values for which one or more most significant bits are decoded (or have been decoded) and which comprise more bits than the most signifi-cant bits, and a second mode in which one or more least significant bits associated with one or more of the spectral values are decoded, while no least significant bits are decod-ed for one or more other spectral values for which one or more most significant bits are decoded (or have been decoded) and which comprise more bits than the one or more most significant bits. Moreover, the audio decoder is configured to provide the decoded audio information using the spectral values.

This embodiment is based on the idea that the first mode or the second mode may be more advantageous in terms of a tradeoff between complexity, bitrate and audio quality depending on the circumstances. The audio decoder can handle two different approaches for dealing with an exhaustion of a bit budget. When operating in the first mode, the audio decoder can handle situations in which an audio encoder omits an encoding of spectral values in the higher frequency range, while spectral values in a low frequency range are all fully encoded (including least significant bits). In the second mode, the audio decoder handles an encoded audio information in which least significant bits are selectively omit-ted for some of the spectral values, even though the one or more most significant bits are encoded for all spectral values. As already mentioned above, both approaches have their advantages depending on some other system parameters (like, for example, the available bitrate), and the audio decoder described here can therefore provide good results under varying conditions.

This audio decoder can also be supplemented by any of the features and functionalities of the above mentioned audio decoder.

In a preferred embodiment, the audio decoder is configured to obtain intermediate bits, bit positions of which are between the least significant bit and the one or more most signifi-cant bits, and the least significant bit associated with a given spectral value from a contig-uous bit sequence in the first mode. Moreover, the audio decoder is configured to obtain intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and the least significant bit associated with a given spectral value from a separate bit sequence or from separate, non-contiguous bit locations of a bit sequence in the second mode.

In other words, in the first mode, there may be a single contiguous bit sequence which encodes both the intermediate bits (as far as intermediate bits are present) and the least significant bits. This contiguous bit sequence, which comprises both the information about the intermediate bits and the information about the least significant bits (but which typically does not comprise information about the one or more most significant bits) can easily be shortened in the case that a bitrate budget is reduced. On the other hand, in the second mode, information representing the least significant bits and the information representing the intermediate bits are contained in separate bit sequences or in separate subsequenc-es of a bit sequence. Accordingly, there is one bit sequence which obtains the information about the intermediate bits (and, optionally, sign information), and there is one sequence which comprises information about the least significant bits (and, optionally, about the signs of values which are very close to zero). Consequently, since the information about the least significant bits is in a separate sequence when operating in the second mode, it is easy to remove or to shorten the sequence comprising the least significant bits, to thereby reduce the bitrate required. The audio decoder can easily adapt to a varying length of the sequence comprising the least significant bits in that the least-significant bit refinement of spectral values is applied to more or less spectral values, depending on how many bits are contained in the sequence representing the least significant bits.

An embodiment according to the invention creates an audio encoder for providing an en-coded audio information on the basis of an input audio information. The audio encoder is configured to obtain spectral values representing an audio content of the input audio in-formation. The audio encoder is also configured to encode at least a plurality of the spec- tral values, in order to obtain an encoded information representing the spectral values (which may be a part of the encoded audio information). Moreover, the audio encoder is configured to jointly encode two or more most significant bits per spectral value, to obtain respective symbol codes for a set of spectral values using an arithmetic encoding. A re-spective symbol code may represent two or more most significant bits per spectral value for one or more spectral values.

The audio decoder is also configured to encode one or more least significant bits associ-ated with one or more of the spectral values in dependence on a bit budget available, such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which two or more most significant bits are encoded and which com-prise more bits than the two or more most significant bits. Moreover, the audio encoder is configured to provide the encoded audio information using the encoded information repre-senting the spectral values.

This audio encoder is based on the idea that a good tradeoff between complexity, bitrate and audio quality can be achieved by selectively omitting an encoding of one or more least significant bits for spectral values for which two or more most significant bits are en-coded using an arithmetic encoding. It has been found that omitting the encoding of one or more least significant bits is not particularly detrimental in the case that there are at least two most significant bits which are encoded.

In particular, it has been found that omitting the encoding of the least significant bits for one or more (quantized) spectral values for which most significant bits are encoded caus-es a much smaller degradation of an audio quality when compared to totally omitting an encoding of some spectral values to remain within a bit budget.

In a preferred embodiment, the arithmetic encoding is configured to determine bit posi-tions (for example, bit weights) of the at least two most significant bits (for example, num-bits and numbits-1), for example, individually for different symbols of the arithmetically encoded representation, and to include into the arithmetically encoded representation an information, for example, an escape sequence comprising one or more "VAL_ESC" sym-bols, describing the bit positions. Accordingly, the bit positions or bit weights of the two or more most significant bits can be adapted to the actual spectral values, wherein the most significant bits can have a large bit weight for comparatively large spectral values and

wherein the most significant bits may have a comparatively small bit weight for compara-tively smaller spectral values. Accordingly, some quantized spectral values may be entire-ly encoded using the two or more most significant bits, wherein there are no least signifi-cant bits (or intermediate bits) remaining. In contrast, other, comparatively larger spectral values may be encoded using two or more most significant bits and using at least one least significant bit. For such comparatively large spectral values for which there is at least one least significant bit, in addition to the two or more most significant bits, the encoder can flexibly decide whether to encode the least significant bit or not, depending on wheth-er an available bit budget is exhausted or not. However, the higher a quantization resolu-tion, the higher the number of spectral values which comprise one or more least signifi-cant bits, in addition to the two or more most significant bits. Accordingly, a possibility for saving bits by not encoding the least significant bits is particularly high for fine quantiza-tion.

In a preferred embodiment, the audio encoder is configured to map at least two most sig-nificant bits of the at least one spectral value onto one symbol of the arithmetically encod-ed representation, which represents the at least two most significant bits of the at least one spectral value. Jointly encoding two or more most significant bits using one symbol of an arithmetically encoded representation has been found to be particularly efficient, since correlations between most significant bits of adjacent spectral values can be exploited, for example when determining a context for the arithmetic encoding.

In a preferred embodiment, the audio encoder is configured to encode, for all spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits and the least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the two or more most significant bits. Accordingly, all the spectral values for which two or more most significant bits are encoded are actually encoded with a good resolution. For such spectral values, all bits except for the least significant bit are always encoded, which brings along a good resolution and has the effect that only the least significant bits are affected in case that a bit budget is exhausted. Thus, a very good hearing impression can be maintained.

In a preferred embodiment, the audio encoder is configured to encode, in a first encoding phase, two or more most significant bits per spectral values and to also encode, in the first encoding phase, for spectral values for which two or more most significant bits are encod-ed and which comprise more bits than the two or more most significant bits (which are

jointly encoded) and a least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the two or more most significant bits. Moreover, the encoder is configured to encode, in the first encoding phase, signs for all spectral values for which two or more most significant bits are encoded and for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value. However, the audio encoder is configured to selective-ly omit, in the first encoding phase, an encoding of a sign for spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value. Accordingly, in the first encoding phase, the most sig-nificant bits and the intermediate bits (as far as intermediate bits are present in between the most significant bits and the least significant bit) are encoded. However, in the first encoding phase, signs are only encoded if the two or more most significant bits and the intermediate bits indicate a non-zero value. In other words, in the first encoding phase, signs are not encoded if the spectral values are so small that they differ from zero only by a least significant bit value (which is the case if the bit weight of the two or more most sig-nificant bits is chosen such that the most significant bits are all zero, which can, for exam-ple, happen if the bit weights of a given spectral value are affected by one or more adja-cent spectral values which are larger than the given spectral value).

Moreover, the audio encoder is configured to selectively encode, in a second encoding phase which follows the encoding phase, sign information for spectral values for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a zero value and for which a least significant bit information indicates a non-zero value. In other words, for very small spectral values, which differ from zero only by a least significant bit value, the sign is only encoded in the second encoding phase, wherein a decision of whether the second encoding phase is actually executed (or com-pleted) for a given spectral value (i.e. whether the least significant bit information is in-cluded into the encoded audio information) is dependent on the bit budget. Thus, the first encoding phase is streamlined, and the sign information is only encoded (e.g. included into the encoded audio information) in the second encoding phase, unless it is already clear from the encoding of the most significant bits and any intermediate bits (as far as there are any intermediate bits) that a sign information is necessary in any case. The en-coding of unnecessary information is avoided and the efficiency is maximized, since it is not clear from the beginning for which spectral values the second encoding phase would be performed. The final decision as to whether the second encoding phase will be per-formed can only be made when it is known how many bits are-needed for the decoding of

the most significant bits and any intermediate bits, and how many bits have already been used by the encoding of other least significant bits.

In a preferred embodiment, the audio encoder is configured to only include a sign infor-mation into the encoded audio representation for spectral values which only differ from zero by a least significant bit if the least significant bit of such spectral values is actually encoded (included in the encoded audio representation). Accordingly, an inclusion of un-necessary information into the encoded audio information (or encoded audio representa-tion) can be avoided. In other words, a sign information is included for all spectral values which are non-zero even when not considering the least significant bit. For spectral values which are non-zero only when considering the least significant bit, the sign information is only included into the encoded audio representation if the least significant bit information is actually included in the encoded audio representation.

In a preferred embodiment, the audio encoder is configured to sequentially provide sub-sequent bits of a least-significant-bit-information bit sequence in order to encode the least significant bit values associated with the spectral values. Accordingly, a contiguous bit sequence or bit stream is provided which only comprises the least significant bit infor-mation and possibly some sign information for such spectral values which are non-zero only when considering the least significant bit. Consequently, there is a separate se-quence of least significant bit information (including associated sign information), which can be shortened or omitted without affecting the encoding of the most significant bits and of the intermediate bits (and of any sign information which is relevant even when leaving the least significant bit unconsidered).

In a preferred embodiment, the audio encoder is configured to provide a single bit of the least-significant-bit-information bit sequence for respective spectral values for which the two or more most significant bit values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the used single bit of the least-significant-bit-information bit sequence is used in order to encode a least significant bit value. Moreover, the audio encoder is configured to provide a single bit of the least-significant-bit-information bit sequence for respective spectral values for which the two or more most significant values and any intermediate bits, as far intermediate bits are pre-sent, indicate a zero value and for which the provided single bit of the least-significant-bit-information bit sequence confirms the zero value. Moreover, the audio encoder is config-ured to provide two subsequent bits of the least-significant-bit-information bit sequence for respective spectral values for which the two or more most significant bits, and any inter-mediate bits, as far as intermediate bits are present, indicate a zero value and for which a first of the provided bits of the least-significant-bit-information bit sequence indicates devi-ation from the zero value by a least significant bit value, wherein a second of the provided bits of the least-significant-bit-information bit sequence encodes a sign of the respective spectral value. In other words, the least-significant-bit-information bit sequence typically comprises one bit per spectral values, but comprises two bits per spectral value if the spectral value deviates from a zero value only by a least significant bit value. In the latter case, the sign information is included into the least-significant-bit-information bit se-quence, because it is only needed if the respective part of the least-significant-bit-information is actually encoded or is actually transmitted to an audio decoder, or is actual-ly evaluated by an audio decoder.

In other words, the sign is selectively included into the least-significant-bit-information bit sequence for spectral values for which the most significant bits and intermediate bits (if present) indicate a zero value and for which the least significant bit indicates a non-zero value (deviating from a zero value only by a least significant bit value).

In a preferred embodiment, the audio encoder is configured to encode the least significant bits starting from a least significant bit associated with a lowest frequency spectral value and proceeding towards spectral values associated with increasingly higher frequencies. Accordingly, encoded information for refining spectral values (for example, for refining all spectral values which comprise more bits than the one or more most significant bits) by least-significant-bit information is provided in a range from a lowest frequency spectral value up to a spectral value for which the "last" least significant bit information is provided. Moreover, no encoded information for refining spectral values by least significant bit in-formation is provided for (all) spectral values (even for encoded spectral values which comprise more bits than the two or more most significant bits) having associated frequen-cies higher than a frequency associated with a spectral value for which the last least sig-nificant bit information is provided. Worded different, unused bits of the bit budget are used for refining spectral values in a low frequency region by a least significant bit infor-mation, until the bit budget is exhausted. Spectral values in a higher frequency region are not refined by the least significant bit information if the bit budget is exhausted. Such a procedure brings along that spectral values in a lower frequency portion are preferred over spectral values in a higher frequency portion when providing least significant bit in-formation. This is in agreement with psycho-acoustic requirements, since a hearing im-

pression will be less distorted by inaccuracies in a higher frequency region when com-pared to inaccuracies in a lower frequency region. Accordingly, the audio encoder can flexibly decide, on the basis of the bit budget, up to which frequency (spectral value for which the last least significant bit information is provided) there is a refinement of spectral values using the least significant bit information in dependence on the bit budget.

In a preferred embodiment, the audio encoder is configured to be switchable between a first mode, in which an encoding of non-zero spectral values in a higher frequency range is (for example, completely) omitted in case that an available bit budget is used up (ex-hausted) by an encoding of spectral values in a lower frequency range and in which least significant bits are encoded for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and a sec-ond mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits.

As already mentioned above, being able to switch between such modes allows for an effi-cient coding in different environments and under different bitrate constraints. In the first mode, the number of spectral values encoded can be varied, and an encoding of non-zero spectral values in a higher frequency range can be omitted in response to an exhaustion of a bit budget. Accordingly, a hearing impression in a high frequency range is degraded, but this may be acceptable under some circumstances, for example in low bitrate envi-ronments. On the other hand, in the second mode, the audio encoder may vary for how many of the spectral values least significant bits are encoded depending on the bit budget, while at least most significant bits are encoded for all spectral values (even in a high fre-quency region). Thus, in the second mode, the encoding precision may be reduced even for lower frequencies in some cases, while there is no total omission of non-zero (quan-tized) spectral values in the high frequency region. The second mode of operation may, for example, result in an improved hearing impression under higher bitrate conditions, which would suffer from a significant degradation if non-zero spectral values in the high frequency region would be totally omitted. Thus, the audio encoder can adapt to different situations and bitrate requirements in a flexible manner by being switchable between the first mode and the second mode.

In a preferred embodiment, the audio encoder is configured to provide a bitstream flag which is included in the encoded audio information (or encoded audio representation) in order to indicate whether the audio encoder operates in the first mode or in the second mode. Accordingly, it is easy for an audio decoder to recognize whether a first decoding mode or a second decoding mode should be used. Using a bitstream flag for such a sig-nal is reasonable, since the audio encoder typically has more knowledge about the specif-ic circumstances than the audio decoder.

In a preferred embodiment, the audio encoder may be configured to jointly encode two or more most significant bits per spectral value for at least two spectral values using respec-tive symbol codes. Accordingly, a respective symbol code may represent two or more most significant bits per spectral value for at least two spectral values. It has been found that such an encoding is particularly efficient, since dependencies and correlations be-tween spectrally adjacent spectral values can be exploited. Also, the bit weight of the most significant bits can be determined on the basis of both spectral values, wherein the spec-tral value having the larger absolute value may decide on the common bit weight of the most significant bits for both spectral values. Accordingly, a signaling overhead for signal-ing the bit weight of the most significant bits can be reduced, because it can be signaled jointly for two or more spectral values.

In a preferred embodiment, the audio encoder is configured to determine an actual high-est-frequency non-zero spectral value (for example, without truncating spectral values) and to encode at least two or more most significant bits of all non-zero (quantized) spec-tral values of all non-zero groups of (quantized) spectral values. Accordingly, it can be ensured that at least most significant bits of all non-zero (quantized) spectral values are encoded, which typically results in a good hearing impression.

In a preferred embodiment, the audio encoder is configured to encode all bits except for a least significant bit for all non-zero (quantized) spectral values. Moreover, the audio en-coder is configured to encode least significant bits for spectral values until a bit budget is exhausted (for example, starting with a lowest frequency spectral value and proceeding towards higher frequency spectral values). Accordingly, a good hearing impression can be achieved and only a variable number of least significant bits will be skipped in the encod-ing, depending on the bit budget.

In a preferred embodiment, the audio encoder is configured to obtain a global gain infor-mation which determines quantization steps of a quantization of spectral values, and which determines a bit demand for the encoding of the quantized spectral values. It has been found that the usage of such a (global) gain information can be helpful to adjust the quantization steps. However, it has also been recognized that it is not easily possible to fine tune a bit demand when using a global gain information. Accordingly, the concept to selectively omit an encoding of least significant bits for some spectral values can be used to compensate for inaccuracies in the adjustment of the bit demand which are caused by the usage of the global gain information. However, it has been found that the combination of the usage of the global gain information with the encoding concept described herein creates a system having a comparatively low computational complexity and still allowing for a good tradeoff between audio quality and bitrate. In particular, a given fixed bitrate can be fully utilized even with a low complexity adjustment of the global gain information by flexibly deciding how many least significant bits should be encoded.

An embodiment according to the invention creates an audio encoder for providing an en-coded audio information on the basis of an input audio information. The audio encoder is configured to obtain spectral values representing an audio content of the input audio in-formation. The audio encoder is configured to encode at least a plurality of the spectral values, in order to obtain an encoded information representing the spectral values. The audio encoder is configured to encode one or more most significant bits using respective symbol codes for a plurality of the spectral values, and to encode one or more least signif-icant bits for one or more of the spectral values, wherein a respective symbol code repre-sents one or more most significant bit values for one or more spectral values. The audio encoder is configured to be switchable between a first mode in which an encoding of non-zero spectral values in a higher frequency range is (for example, entirely) omitted in case that an available bit budget is used up (for example, exhausted) by encoded spectral val-ues in a lower frequency range and in which least significant bits are encoded for all spec-tral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and a second mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the most signif-icant bits. The audio encoder is configured to provide the encoded audio information using the encoded information representing the spectral values.

This audio encoder is based on the considerations mentioned above for the similar audio encoder and also for the similar audio decoder. In particular, while being switchable be-tween the first mode and the second mode, the audio encoder can adapt to different en-coding situations and bitrate requirements.

In a preferred embodiment, the audio encoder is configured to encode one or more most significant bits of all non-zero spectral values or of all non-zero groups of spectral values in the second mode. Accordingly, a good hearing impression can be reached.

In a preferred embodiment, the audio encoder is configured to limit, when operating in the first mode, a frequency range for which the spectral values are encoded, in case a bit budget is insufficient, such that one or more spectral values (for example, in a high fre-quency range) are left unconsidered in the encoding of spectral values. Thus, a selective limitation of the frequency range, in dependence on a bit budget, is used in the first mode, wherein the limitation of the frequency range helps to save bits.

In a preferred embodiment, the audio encoder is configured to determine, when operating in the first mode, a maximum frequency value and to encode, when operating in the first mode, spectral values up to the maximum frequency and to leave, when operating in the first mode, spectral values above the maximum frequency unencoded even if the spectral values are non-zero (or have non-zero most significant bits). Moreover, the audio encoder is configured to select, when operating in the first mode, the maximum frequency value in dependence on a computation or estimation of a bit demand for encoding all spectral val-ues, such that a number of spectral values to be encoded is reduced if the computed or estimated bit demand would exceed a bit budget. Moreover, the audio encoder is config-ured to determine, when operating in the second mode, the maximum frequency value (for example, to be equal to an actual maximum frequency value) and to encode, when oper-ating in the second mode, spectral values up to the maximum frequency and to leave, when operating in the second mode, spectral values above the maximum frequency un-encoded. When operating in the second mode, the maximum frequency value is selected such that at least one or more most significant bits of all non-zero spectral values or of all non-zero groups of spectral values are encoded and such that at most zero-valued spec-tral values are left unencoded. In other words, the audio encoder uses different criteria for the selection of the maximum frequency value in the different modes. In the first mode, the maximum frequency value is chosen in dependence on the bit demand, wherein non-zero (quantized) spectral values are left unencoded in case the bit budget is too small. On the other hand, in second mode, the maximum frequency value is chosen such that for all spectral values quantized to a non-zero value at least the one or more most significant bits are encoded. Thus, different concepts are used for dealing with an exhaustion of a bit budget. In the first mode, an exhaustion of a bit budget is handled by reducing the maxi-mum frequency value. In the second mode, an exhaustion of a bit budget is handled by omitting the encoding of least significant values of one or more spectral values for which the most significant bits are encoded.

In a preferred embodiment, the audio encoder is configured to include an information de-scribing the maximum frequency into the encoded audio information. Accordingly, an au-dio decoder knows how many spectral values should be decoded. The information de-scribing the maximum frequency can be used both for a limitation of the number of en-coded (and decoded) spectral values due to an exhaustion of a bit budget and also to signal that all spectral values above the maximum frequency are zero (e.g. actually zero even without truncation).

In a preferred embodiment, the audio encoder is configured to make a mode decision whether to use the first mode or the second mode in dependence on an available bitrate (for example, such that the first mode is used for comparatively smaller bitrates and such that the second mode is used for comparatively higher bitrates).

Such a mechanism is useful, since the second mode is better suited for dealing with an exhaustion of a bit budget in the case of higher bitrates. In contrast, the first mode some-times brings better results than the second mode in the case of comparatively low bitrates.

In another preferred embodiment, the audio encoder is configured to make a mode deci-sion whether to use the first mode or the second mode in dependence on an information on a number of spectral values or groups of spectral values which comprise, in addition to one or more most significant bits encoded in a most-significant-bit-encoding step, one or more least significant bits, an encoding of which can selectively be omitted in dependence on a bit demand and a bit budget. Such a concept is helpful since the second mode is best suited for cases in which there is (after the quantization) a large number of least sig-nificant bits. Such a large number of least significant bits is, for example, present in the case of a high bitrate, where an encoding can be done with a high resolution (and wherein a fine quantization can be used).

In a preferred embodiment, the audio encoder is configured to include a bitstream flag into the encoded audio information indicating whether the audio encoder operates in the first mode or in the second mode. Accordingly, an audio decoder can be informed which de-coding mode to use.

In a preferred embodiment, the audio encoder is configured to encode intermediate bits, bit positions of which are between the least significant bit and the one or more most signif-icant bits, and a least significant bit associated with a given spectral value into a contigu-ous bit sequence in the first mode. Moreover, the audio encoder is configured to encode intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and the least significant bit associated with a given spectral value into separate bit sequences or into separate, non-contiguous bits locations (or bit-stream portions) of a bit sequence in the second mode. Accordingly, when operating in the first mode, there is a contiguous bit sequence which represents both intermediate bits and the least significant bits. In contrast, when operating in the second mode, intermedi-ate bits and least significant bits are provided in separate sequences (or into separate portions of a common sequence) which allows for a simple shortening of the sequence representing the least significant bits. Accordingly, an adaptation to the bit budget is easily possible, even after the encoding is completed. This facilitates adaptation to the bit budg-et.

In a preferred embodiment, the audio encoder is configured to encode, when operating in the first mode, a sign information associated with a spectral value in a bit sequence which is associated with intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and least significant bits. Moreover, the au-dio encoder is configured to selectively encode, when operating in the second mode, a sign information associated with a spectral value in a bit sequence which is associated with intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, or in a bit sequence associated with least significant bits (and sign information) such that sign information for spectral values which deviate from zero only by a least significant bit value are encoded in the bit sequence associated with least significant bits (and sign information). Accordingly, the sign information is placed within the bit sequence associated with least significant bits (and sign information) in the case that the sign information is only needed when the least significant bit information is evaluated. Accordingly, the information which is always included into the encoded audio representation, namely the bit sequence which is associated with intermediate bits and

sign information, does not comprise any information which is unnecessary in the case that the least significant bit information is omitted. This simplifies scalability of the bitrate.

An embodiment according to the invention creates an audio decoder for providing an en-coded audio information on the basis of an input audio information. The audio encoder is configured to obtain spectral values representing an audio content of the input audio in-formation (for example, using a MDCT transform). The audio encoder is configured to encode at least a plurality of the spectral values, in order to obtain an encoded information representing the spectral values. The audio encoder is configured to obtain a (global) gain information which determines quantization steps of a quantization of spectral values, and which determines a bit demand for encoding the quantized spectral values. The audio encoder is configured to encode one or more most significant bits using respective symbol codes for a plurality of the spectral values using an arithmetic encoding, and to encode one or more least significant bits for one or more of the spectral values, wherein a respec-tive symbol code represents one or more most significant bits per spectral value for one or more spectral values. The audio encoder is configured to encode one or more least signif-icant bits associated with one or more of the spectral values in dependence on a bit budg-et available, such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the one or more most significant bits. Moreover, the audio en-coder is configured to provide the encoded audio information using the encoded infor-mation representing the spectral values.

This audio encoder is based on the finding that the usage of a gain information (or global gain information) is useful for defining a quantization. Also, the concept to selectively en-code least significant bits is very efficient in combination with this concept. For details, reference is also made to the discussion above.

In a preferred embodiment, the audio encoder is configured to obtain a first estimate of the gain information based on an energy of groups of spectral values (for example, MDCT coefficients). Moreover, the audio encoder is configured to quantize the set of spectral values (for example, a MDCT spectrum) using the first estimate of the gain information. Moreover, the audio encoder is configured to compute or estimate a number of bits need-ed to encode the set of spectral values quantized using the first estimate of the gain in-formation or using a refined gain information. Moreover, the audio encoder is configured to decide whether to use the first mode or the second mode in dependence on a number of bits needed. Accordingly, a decision about the quantization, and also a decision which mode to use, can be made in an efficient way. Depending on whether an iterative proce-dure is to be chosen or not, the number of bits needed to encode the set of spectral val-ues can be estimated using a quantization in dependence on the first estimate of the gain information or using a quantization in dependence on an iteratively refined gain infor-mation. Accordingly, the complexity of the determination of the quantization accuracy can be kept reasonably small.

In a preferred embodiment, the audio encoder is configured to be switchable between the first mode and the second mode mentioned above. In particular, the audio encoder is con-figured to decide whether to use the first mode or the second mode in dependence on a number of bits needed and in dependence on a criterion which indicates how many spec-tral values comprise more bits than the one or more most significant bits. In particular, the number of bits needed, which can be determined after deciding on the gain information to be used (first estimate or refined gain information) can be compared with a bit budget, and the decision which mode to use can made both in dependence on this comparison and in dependence on the criterion which indicates how many spectral values comprise more bits than the one or more most significant bits. Accordingly, the second mode can be used if there are many spectral values comprising one or more less significant bits in addition to the one or more most significant bits.

In a preferred embodiment, the audio encoder is configured to be switchable between the first mode and the second mode mentioned above. In this case, the audio encoder can be configured to decide whether to use the first mode or the second mode in dependence on the number of bits needed and in dependence on a bitrate, such that the second mode is chosen if the bitrate is larger than or equal to a threshold bitrate and if a computed or es-timated number of bits needed to encode the set of spectral values is higher than a bit budget. It has been shown that the usage of the second mode is particularly helpful in said case.

Moreover, the audio encoder can also be supplemented by any of the other features men-tioned before. The same advantages discussed before also apply.

Further embodiments according to the invention create methods for providing a decoded audio information on the basis of an encoded audio information and methods for providing an encoded audio information on the basis of an input audio information. These methods correspond to the respective audio decoder and to the respective audio encoder and can be supplemented by any of the features and functionalities discussed herein with respect to the corresponding audio decoder or audio encoder.

Further embodiments according to the invention comprise a computer program for per-forming any of the methods described herein.

A further embodiment comprises a bitstream, which is based on the same considerations discussed above and which may be supplemented by any of the information items to be encoded and decoded as mentioned herein.

Brief Description of the Figures

Embodiments according to the present invention will subsequently be described taking reference to the enclosed Figures, in which:

Fig. 1 shows a block schematic diagram of an audio decoder, according to an embodi-ment of the present invention;

Fig. 2 shows a block schematic diagram of an audio decoder, according to another em-bodiment of the present invention;

Fig. 3 shows a block schematic diagram of an audio encoder, according to an embodi-ment of the present invention;

Fig. 4 shows a block schematic diagram of an audio encoder, according to an embodi-ment of the present invention;

Fig. 5 shows a block schematic diagram of an audio encoder, according to an embodi-ment of the present invention;

Fig. 6 shows a block schematic diagram of another audio encoder, according to an em-bodiment of the present invention;

Fig. 7 shows a block schematic diagram of an audio decoder according to another embod-iment of the present invention;

Fig. 8 shows a flowchart of a functionality of an audio encoder, according to an embodi-ment of the present invention;

Fig. 9 shows a flowchart of a functionality of an audio decoder, according to an embodi-ment of the present invention;

Figs. 10a-10f show pseudo program code representations of functionalities of an audio encoder, according to an embodiment of the present invention.

Figs. 11a-11d show pseudo program code representations of the functionalities of an au-dio decoder, according to an embodiment of the present invention;

Fig. 12 shows a graphic representation of a signal-to-noise ratio generated by a conven-tional audio encoder/decoder;

Fig. 13 shows a graphic representation of a signal-to-noise ratio provided by audio encod-ers/decoders according to the present invention; and

Figs. 14-18 show flowcharts of methods for audio encoding and audio decoding, accord-ing to embodiments of the present invention.

1). Audio Decoder According to Fig. 1

Fig. 1 shows a block schematic diagram of an audio decoder 100 according to an embod-iment of the present invention.

The audio decoder 100 is configured to recieve an encoded audio information 110 and to provide, on the basis thereof, a decoded audio information 112. The audio decoder 100 is configured to obtain decoded spectral values 132 on the basis of an encoded information 130 representing the spectral values, wherein the encoded information 130 may be part of the encoded audio information 110. In addition, the encoded audio information 110 may optionally comprise further information, like noise shaping information, control information and the like.

The audio decoder is configured to jointly decode two or more most significant bits per spectral value (for example, per quantized spectral value) on the basis of respective sym-bol codes (for example, symbol codes of an arithmetically encoded representation of the most significant bits) for a set of spectral values using an arithmetic decoding. A respec-tive symbol code may represent two or more most significant bits per spectral value for one or more spectral values. The arithmetically encoded symbol codes may, for example, be part of the encoded information 130 representing spectral values.

Moreover, the audio decoder is configured to decode one or more least significant bits associated with one or more spectral values in dependence on how much least significant bit information is available. The least significant bit information, which can be considered as a representation of least significant bits, may also be part of the encoded information 130 representing spectral values.

In particular, the audio decoder may be configured to decode one or more least signifi-cant bits associated with one or more of the spectral values in dependence on how much least significant bit information is available, such that one or more least significant bits associated with one or more of the (quantized) spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits are decoded (or have been decoded) and which comprise more bits than the one or more most significant bits.

In other words, the audio decoder may be configured to decode least significant bits for some of the spectral values for which two or more most significant bits have been decod-ed, and the audio decoder may omit a decoding of one or more least significant bits for some other spectral values for which one or more most significant bits have been decod-ed.

Wording it yet differently, the audio decoder may, for example, only refine a true subset of the spectral values, for which most significant bits have been decoded, wherein the num-ber how many spectral values are refined by least significant bits depends on how much least significant bit information is available (for example, how much least significant bit information is included in the encoded audio information 110 by an audio decoder in view of bit budget constraints).

The audio decoder 100 can optionally be supplemented by any of the features, functionali-ties and details described herein, either individually or in combination.

2). Audio Decoder According to Fig. 2

Fig. 2 shows a block schematic diagram of an audio decoder 200, according to an embod-iment of the present invention.

The audio decoder 200 is configured to receive and encoded audio information 210 and to provide, on the basis thereof, a decoded audio information 212.

The encoded audio information 210 may, for example, comprise an encoded information 230 representing spectral values, wherein the encoded information 230 representing spectral values may, for example, comprise arithmetically encoded symbol codes repre-senting one or more most significant bits and a representation of least significant bits ond of signs. The encoded audio information 210 may optionally comprise further information, like for example, control information of noise shaping information. The optional further information may also be used in the decoding process, but is not essential for the present invention.

The audio decoder is configured to obtain decoded spectral values 232 on the basis of the encoded information 230 representing the spectral values.

The audio decoder is configured to decode one or more most significant bits on the basis of respective symbol codes (for example, on the basis of arithmetically encoded symbol codes) for a plurality of spectral values, and to decode one or more least significant bits for one or more of the spectral values. For example, the audio decoder may use the arithmetically encoded symbol codes and the representation of least significant bits, which may be included in the encoded information 130.

The audio decoder 200 is configured to be switchable between a first mode in which a decoding of spectral values in a higher frequency range is omitted (for example, entirely omitted) in response to a signaling from an encoder and in which least significant bits are decoded for all spectral values for which one or more most significant bits have been de-coded and which comprise more bits than the most significant bits, and a second mode in which one or more least significant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits have been decoded and which com-prise more bits than the most significant bits. In other words, in the first mode, the audio decoder 200 may, for example, decode only spectral values in a lower frequency range (for example, up to a frequency determined and signaled by an audio encoder) while omit-ting a decoding of spectral values in a higher frequency range (for example, above the frequency specified by the encoder). However, in the first mode, a full number representa-tion of the spectral values may be decoded for all spectral values in the lower frequency range, such that most significant bits, any intermediate bits and any least significant bits are decoded for all spectral values in the lower frequency range. In contrast, in the second mode, the audio decoder may only decode least significant bits for some of the spectral values for which one or more most significant bits are decoded, but not for all spectral values for which one or more most significant bits are decoded. Thus, in the second mode, least significant bits may be decoded in one frequency region but not in another frequency region (for example, in a higher frequency region).

Moreover, the audio decoder 200 is configured to provide decoded audio information 212 using the spectral values 232. For example, the audio decoder 200 may comprise a fur-ther processing of the decoded spectral values 232, which, details of which, however, are not of particular relevance for the subject-matter of the present invention.

Moreover, it should be noted that the audio decoder 200 can be supplemented by any of the features, functionalities and details described herein, either individually or in combina-tion.

3). Audio Encoder According to Fig. 3

Fig. 3 shows a block schematic diagram of an audio encoder 300 according to an embod-iment of the present invention. The audio encoder 300 is configured to receive an input audio information 310 and may provide an encoded audio information 312 (which may correspond to the encoded audio information 110, 210). The audio encoder 300 is config-

ured to obtain spectral values 330 representing an audio content of the input audio infor-mation 310. For example, the audio decoder 300 may optionally comprise any form of preprocessing, like, for example, a time-domain-to-spectral-domain conversion (for exam-ple, an MDCT) ans/or a spectral shaping (in the time domain and/or in the spectral do-main) in order to obtain the spectral values 330.

The spectral values 330 may, for example, be quantized (preferably integer) values in a signed binary number representation. Moreover, the audio encoder is configured to en-code at least a plurality of the spectral values 330, in order to obtain an encoded infor-mation 350 representing the spectral values 330. The audio encoder 300 may, for exam-ple, be configured to provide the encoded audio information 312 using the encoded infor-mation 350 representing the spectral values. However, the audio encoder 300 may op-tionally also provide further information, like control information or noise shaping infor-mation, which is also included in the encoded audio information 312 (but details of which are not of particular relevance for the present invention).

The audio encoder 300 is configured to jointly encode two or more most significant bits per spectral value, to obtain respective symbol codes for a set of spectral values using an arithmetic encoding. A respective symbol code may, for example, represent two or more most significant bits per spectral value for one or more spectral values.

The audio encoder is further configured to encode one or more least significant bits asso-ciated with one or more of the spectral values 330 in dependence on a bit budget, such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits.

For example, the audio encoder 300 may only provide encoded least significant bits for spectral values in a lower frequency portion but not for spectral values in a higher fre-quency portion. By selecting for which spectral values least significant bits are provided, a number of bits can be adapted to a bit budget.

Moreover, it should be noted that the audio encoder according to Fig. 3 can be supple-mented using any of the features, functionalities and details described herein, either indi-vidually or in combination.

4). Audio Encoder According to Fig. 4

Fig. 4 shows a block schematic diagram of an audio encoder 400, according to an embod-iment of the present invention.

The audio encoder 400 is configured to receive an input audio information 410 and to pro-vide, on the basis thereof, an encoded audio information 412. The audio encoder is con-figured to obtain spectral values 330 (which may, for example, be quantized (preferably integer) spectral values in a signed binary number representation) representing an audio content of the input audio information 410. For example, an optional preprocessing may be used, which may, for example, comprise a time-domain to frequency-domain conver-sion and/or a noise shaping. Moreover, a quantization may optionally be used to obtain the spectral values 430.

The audio encoder is further configured to encode at least a plurality of the spectral values 430, in order to obtain an encoded information 450 representing the spectral values. The audio encoder is configured to encode one or more most significant bits (of the spectral values) using respective symbol codes for a plurality of spectral values and to encode one or more least significant bits for one or more of the spectral values. A respective symbol code may, for example, represent one or more most significant bit values for one or more spectral values. The audio encoder may be configured to be switchable between a first mode, in which an encoding of non-zero spectral values in a higher frequency range is omitted (for example, entirely omitted) in case that an available bit budget is used up (ex-hausted) by an encoding of spectral values in a lower frequency range and in which least significant bits are encoded for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and a sec-ond mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the one or more most significant bits.

In other words, the audio encoder may, for example, encode a comparatively smaller number (for example, not all non-zero spectral values) in the first mode, but those spectral values which are encoded are encoded with full accuracy (including the least significant bit). In contrast, in the second mode, the audio encoder may, for example, encode at least the most significant bits of all non-zero spectral values, but may encode some of the spec-tral values with reduced resolution (for example, without encoding the corresponding least significant bit). Thus, the encoder may, for example, be switchable between two modes which provide different mechanisms for adapting a number of bits to the bit budget, wherein the first mode relies on an omission of an encoding of spectral values in a higher frequency range for the reduction of the number of bits, and wherein the second mode relies on an omission of least significant bits for some spectral values (for which only the most significant bit and possibly some intermediate bits are encoded, and which are "par-tially encoded").

The audio encoder 400 according to Fig. 4 can be supplemented by any features, func-tionalities and details described herein, either individually or in combination.

5). Audio Encoder According to Fig. 5

Fig. 5 shows a block schematic diagram of an audio encoder 500, according to an embod-iment of the present invention. The audio encoder 500 is configured to receive an input audio information 510 and to provide, on the basis thereof, an encoded audio information 512. The audio encoder is configured to obtain spectral values 530 representing an audio content of the input audio information 510. For example, the audio encoder may use a modified discrete cosine transform (MDCT) to obtain the spectral values 530. Generally speaking, the audio encoder 500 may optionally use any type of preprocessing, like a time-domain-to-frequency-domain conversion and a noise shaping, and the audio encoder 500 may optionally also use a quantization. For example, the spectral values 530 may be quantized spectral values or may be noise-shaped and quantized MDCT coefficients.

The audio encoder is configured to encode at least a plurality of the spectral values 530, in order to obtain an encoded information 550 representing the spectral values. The en-coded information 550 may be a part of the encoded audio information 512. However, the encoded audio information 512 may also comprise, optionally, further information, like a control information or a spectral shaping information.

The audio encoder 500 is also configured to obtain a gain information (for example, a global gain information) 560, which determines quantization steps of a quantization of spectral values, and which determines a bit demand for encoding the quantized spectral values.

The audio encoder 500 is configured to encode one or more most significant bits (of the quantized spectral values) using respective symbol codes for a plurality of the (quantized) spectral values using an arithmetic encoding, and to encode one or more least significant bits for one or more of the (quantized) spectral values. A respective symbol code may, for example, represent one or more most significant bits per spectral value for one or more spectral values.

The audio encoder is configured to encode one or more least significant bits associated with one or more of the (quantized) spectral values in dependence on a bit budget availa-ble, such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which com-prise more bits than the one or more most significant bits. For example, the audio encoder may only provide encoded least significant bits for some of the spectral values, while no least significant bit information is provided for other spectral values which would also ben-efit from a least significant bit refinement.

Moreover, the audio encoder 500 is configured to provide the encoded audio information 512 using the encoded information 550 representing the spectral values.

It should be noted that the audio encoder 500 can be supplemented by any of the fea-tures, functionalities and details described herein, either individually or in combination.

6). Audio Encoder According to Fig. 6

Fig. 6 shows a block schematic diagram of an audio encoder, according to an embodi-ment of the invention.

The audio encoder according to Fig. 6 is designated in its entirety with 600.

The audio encoder 600 is configured to receive an input audio information 610 and to pro-vide, on the basis thereof, an encoded audio representation 612.

The audio encoder 600 may comprise an optional preprocessing 620, which may apply some type of preprocessing (like, for example, a filtering, a bandwidth limitation, a time-domain noise shaping, or the like) on the input audio signal.

The audio encoder 600 may optionally comprise a time-domain-to-spectral-domain con-version 630 which may, for example, perform a modified-discrete-cosine-transform or a similar transform, like a low-delay-modified-discrete-cosine-transform. The time-domain-to-spectral-domain conversion 630 may, for example, receive the input audio information 610, or the preprocessed version 622 thereof and provide spectral values 632.

The audio encoder 600 may optionally comprise a (further) preprocessing, which receives the spectral values 632 and which may, for example, perform a noise shaping. For exam-ple, the (further) preprocessing 640 may perform a spectral noise shaping and/or a tem-poral noise shaping. Optionally, the preprocessing 640 may, for example, apply scale fac-tors to scale different frequency bands ("scale factor bands") in accordance with a psy-choacoustic relevance of the frequency bands (which may be determined, for example, by a psychoacoustic model). Accordingly, preprocessed spectral values 642 may be ob-tained.

The audio encoder 600 may optionally comprise a scaling 650 which may, for example, scale the spectral values 632 or the preprocessed spectral values 642. For example, the scaling 650 may scale the spectral values 632 or the preprocessed spectral values 642 using a global gain, to thereby provide scaled spectral values 652.

The audio encoder 600 also comprises a quantization (or quantizer) 660, which may re-ceive the spectral values 632, the preprocessed spectral values 642 or the scaled spectral values 652. The quantization 660 may, for example, quantize the spectral values 632 or the preprocessed spectral values 642 or the scaled spectral values 652, to thereby obtain quantized spectral values 662, which may, for example, be signed integer values and which may, for example, be represented in a binary representation (for example, in a two's-complement representation). The quantized spectral values 662 may, for example, be designated with Xq. For example, a predetermined number of 256, 512, 1024 or 2048 quantized spectral values may be provided per frame, wherein different frequencies are associated with the quantized spectral values.

The encoder 600 may also comprise an encoding 670, which receives the quantized spectral values 662 (Xq) and which may provide, on the basis thereof, an encoded infor-mation representing the (quantized) spectral values 672.

It should be noted that the quantized spectral values 662 may correspond to the spectral values 330, 430, 530 and that the encoded information 672 representing spectral values may correspond to the encoded information 350, 450, 550 representing spectral values. Moreover, it should be noted that the encoding 670 may, for example, perform the func-tionalities described with respect to the encoders 300, 400, 500. However, the encoding 670 may also comprise the functionality described in the following (for example with refer-ence to Fig. 8), or at least part of said functionality.

The audio encoder 600 also optionally comprises a post processing 680, which may apply a post processing to the encoded information 672.

Accordingly, the encoded representation 612 is provided, which typically comprises the encoded information 672. However, the encoded audio representation 612 may optionally also comprise additional information, like control information and information regarding the noise shaping (like scale factor information, linear prediction coefficients, or the like). The encoded audio representation may optionally also comprise global gain information and/or encoding mode information/decoding mode information and/or a "lastnz" information.

To conclude, the concept for the encoding of spectral values disclosed herein can, for example, be implemented in an audio encoder 600, wherein only some or all of the fea-tures of the scale factor encoding described herein can be taken over in the audio encod-er 600.

7). Audio decoder according to Fig. 7

Fig. 7 shows a block schematic diagram of an audio decoder 700, according to an embod-iment of the present invention. The audio decoder 700 is configured to receive an encod-ed audio information 710 (which may, for example, correspond to the encoded audio rep-resentation 612) and may provide, on the basis thereof, a decoded audio information 712. The audio encoder 700 may, for example, comprise a decoding 720 which receives the encoded audio information, or a part thereof, and provides, on the basis thereof, quan-tized spectral values 722 (also designated with Xq). For example, the decoding 720 may

provide signed integer values in a binary representation (for example, in a two's comple-ment representation).

The audio decoder 700 optionally comprises an inverse quantizer 730, which receives the quantized spectral values and which may perform an inverse quantization. For example, the inverse quantizer 730 may use a global gain information to adjust the mapping per-formed by the inverse quantization.

The audio decoder 700 optionally comprises a scaling 740, which may receive inversely quantized spectral values 732 provided by the inverse quantizer and which may perform a scaling, to thereby obtain inversely quantized and scaled spectral values 742. The scaling may optionally be dependent on the global gain.

The audio decoder 700 may also, optionally, comprise a post processing 750, which may receive the inversely quantized spectral values 730 or the inversely quantized and scaled spectral values 742 and which may perform a spectral shaping. For example, the spectral shaping may be a spectral noise shaping, and/or may be based on a scaling of different frequency bands using scale factors and/or may be based on a spectral shaping using linear prediction coefficients (wherein information controlling the spectral shaping may be included in the encoded audio information).

The audio decoder 700 may also, optionally, comprise a spectral-domain-to-time-domain conversion 760, which may receive the inversely quantized spectral values 732, the in-versely quantized and scaled spectral values 742 or the post processed (for example, spectrally shaped) spectral values 752 provided by the post processing 750. The spectral-domain-to-time-domain conversion may, for example, perform an inverse modified dis-crete cosine transform, or a low-delay-inverse-modified-discrete-cosine-transform, or any other spectral-domain-to-time-domain conversion, to thereby obtain a time-domain audio representation 762 on the basis of the input information received by the spectral-domain-to-time-domain conversion.

The time-domain audio representation 762 may, for example, be input into an (optional) post processing 770, which may perform one or more post processing steps and which may, for example, also perform a time domain spectral shaping (for example, in the case that no spectral shaping is performed in the spectral domain, for example, using an LPC filtering).

Accordingly, the decoded audio information 712 may be provided on the basis of the out-put of the spectral-domain-to-time-domain conversion 762, and may possibly be obtained using some form of post processing and/or frame linking (like an overlap-and-add opera-tion).

To conclude, the audio decoder 700 may perform some audio decoding functionality, wherein details, for example regarding a noise shaping or spectral shaping, may vary sig-nificantly from implementation to implementation. The spectral shaping or noise shaping may be performed in the spectral-domain (i.e., before the spectral-domain-to-time-domain conversion) and/or in the time-domain (for example, after the spectral-domain-to-time-domain conversion).

However, it should be noted that the encoded audio information 710 may correspond to the encoded audio information 110, 210, and that the encoded audio information 710 may comprise additional control information and information for adjusting a spectral shaping. Moreover, the quantized spectral values 722 may, for example, correspond to the decod-ed spectral values 132, 232.

Also, the decoding 720 may perform some or all of the functionalities described with re-spect to the audio decoders 100, 200.

Also, the decoding 720 may be supplemented by any of the features, functionalities and details described herein with respect to a decoding of spectral values (or spectral coeffi-cients) which is disclosed herein, either individually or in combination.

8). Audio encoding (spectral value encoding) according to Fig. 8

Fig. 8 shows a flowchart of a functionality which may be performed by any of the audio encoders described herein.

It should be noted that some or all of the functionalities described with reference to Fig. 8 (and also with respect to the following figures) can be taken over into the audio encoders of Figs. 3, 4, 5 and 6.

It should also be noted that Fig. 8 is focused on the encoding of spectral values, which may, typically, be quantized spectral values. Preferably, but not necessarily, the spectral values are signed integer values, which are represented by a binary two's-complement representation.

The functionality as shown in the flowchart 800 comprises a first estimation 810 of a glob-al gain. This estimation may, for example, be made on the basis of a set of spectral val-ues, which may be associated with a frame of an audio content, and may also consider a bit budget (or, equivalently, a bitrate budget).

The functionality of the audio encoder or of the audio encoding, as shown in Fig. 8, also comprises a quantization 814 of spectral coefficients (or, equivalently, spectral values) using a first estimate of a global gain or using a refined estimate of a global gain (which may be obtained in an iterative manner). In the step 814, there is a computation or estima-tion of a number of bits needed to encode the quantized spectrum (which may be repre-sented by quantized spectral coefficients or by quantized spectral values).

On the basis of this computation or estimation of the number of bits needed, which is per-formed in step 818, the global gain may optionally be adjusted or refined in a step 822, to thereby obtain an improved estimate of the global gain.

Accordingly, by performing steps 810, 814 and 818, and optionally step 822, a "global gain information" (or, generally, an information describing a quantization of the spectral values) may be obtained, which results in a quantization such that an expected number of bits at least approximately fits a bit budget. However, it should be noted that, in view of complexity constraints, the global gain information may not be quite appropriate, such that an encoding of spectral values quantized in dependence on the global gain information may still consume more or less bits when compared to the bit budget.

It should be noted that any details regarding the computation of the global gain or regard-ing the quantization are not essential for the present invention. Rather, embodiments ac-cording to the present invention will work with any mechanism which provides quantized spectral values such that the spectral values can be encoded without excessively violating the bit budget.

The functionality 800 further comprises performing a mode decision 830. However, per-forming the mode decision can be considered as optional, since audio encoders using only one mode (designated herein as "second mode") are also possible. The mode deci-sion 830 optionally comprises a mode-dependent identification of a last encoded coeffi-cient. Depending on the mode decision, the determination of the last encoded coefficient may be performed in a different manner.

If the "first mode" is used, there may be a decision not to encode some non-zero spectral values in order to save bits (and to stay within the bit budget). In this case, a frequency associated with the last encoded spectral coefficient may be chosen to be smaller than a maximum frequency at which there is a non-zero spectral value. Consequently, some non-zero spectral values in a high frequency region may not be encoded in the first mode.

In contrast, in the second mode, at least most significant bits are encoded for all non-zero spectral coefficients. Accordingly, the last encoded coefficient may, for example, be cho-sen to be the highest-frequency non-zero spectral value.

An index describing the highest frequency spectral value which will be encoded is provid-ed as a control information "lastnz" both in the first mode and in the second mode.

In the following, operation in the "first mode" will be described, taking reference to func-tionalities 840 to 869.

Operation in the first mode comprises an arithmetic encoder initialization 840. In this step, states and a context of the arithmetic encoder will be initialized.

In the step 844, some side information will be encoded, like a mode information indicating the usage of the first mode, the global gain and the information identifying the last encod-ed coefficient (lastnz).

Functionalities 848 to 864 are repeated for each spectral value, or for each group of spec-tral values. It should be noted that, in a preferred embodiment, groups comprising two spectral values will be encoded. However, an encoding of individual spectral values is also possible.

The actual encoding of spectral values comprises a most significant bit weight determina-tion for a spectral coefficient or for a group of spectral coefficients. For example, the num-ber representation of one or two spectral coefficients is examined and it is identified which is the highest-valued bit position which comprises a "1". For example, a binary value "00010000" comprises its most significant bit at bit position 5, having bit weight 16. If a pair of spectral values is considered, which is to be encoded together, the maximum of the most significant-bit positions of the two spectral values is determined. For optional details, reference is made to the description of "step 7a" which will be provided below (confer the description of Fig. 10a).

In a step 852, a most significant bit weight will be encoded, which can be done, for exam-ple, by providing a sequence of specific encoded symbols, wherein the number of specific encoded symbols indicates the bit position (or, equivalently, the bit weight). For example, so-called "escape-symbols" can be used, which are known in arithmetic coding. For op-tional details regarding functionality 852, reference is made, for example, to the discussion of "step 7b" provided below (confer Fig. 10c).

Subsequently, a most significant bit encoding 856 is performed. In this step, one or more bits (for example, two bits) at the identified most-significant-bit bit position (or adjacent to the identified most-significant-bit bit position) are encoded. For example, if bit position 5, having bit weight 16 is identified in step 848, then bits having bit positions 5 and 4 (bit weights 16 and 8) of a first spectral value may be encoded together with bits at bit posi-tions 5 and 4 (having bit weights 16 and 8) of a second spectral value. Thus, in the exam-ple, a total of four bits may be encoded together, wherein typically at least one of the two spectral values will have a "1" at bit position 5 (bit weight 16). For example, the four men-tioned bits may be mapped onto a symbol "sym" using a context-based arithmetic coding. For optional details, reference is made, for example, to the description of "step 7c" which is provided below (confer Fig. 10d).

In a step 860, there is a remaining bit encoding. In the step 860, there may, for example, be an encoding of (all) less significant bits (including one or more least significant bits) for all spectral values which have been encoded in step 856 and which comprise more bits than the one or more most significant bits (e.g. numbits>2). In other words, for each spec-tral value which has partially (but not completely) been encoded in step 856 (because the encoding of the one or more most significant bits was not sufficient to represent the spec- tral value with full accuracy, down to a bit having bit weight 1), all less significant bits will be encoded.

Taking reference to the above example, if bits 5 and 4 have been encoded in step 856 for the first spectral value and for the second spectral value, then bits 1, 2 and 3 will be en-coded for the first spectral value and for the second spectral value in step 860.

Claims

1. An audio decoder (100; 200;700) for providing a decoded audio information (112;

212; 712) on the basis of an encoded audio information (110;210;710), wherein the audio decoder is configured to obtain decoded spectral values (132;232;732; Xq[n], Xq[n+1]) on the basis of an encoded information (130; 230) representing the spectral values,

wherein the audio decoder is configured to jointly decode (950; 1110a-1110g) two or more most significant bits per spectral value on the basis of respective symbol codes (sym) for a set of spectral values using an arithmetic decoding,

wherein a respective symbol code (sym) represents two or more most significant bits per spectral value for one or more spectral values,

wherein the audio decoder is configured to decode (972; 1140a-1141j) one or more least significant bits associated with one or more of the spectral values in dependence on how much least significant bit information is available,

such that one or more least significant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits are decoded and which comprise more bits than the one or more most significant bits; and

wherein the audio decoder is configured to provide the decoded audio information using the spectral values.

2. The audio decoder according to claim 1, wherein the audio decoder is configured to map (1110f, 1110g) one symbol (sym) of an arithmetically encoded representa- tion, which represents at least two most significant bits of at least one spectral val- ue, onto the at least two most significant bits of the at least one spectral value.

3. The audio decoder according to claim 1 or claim 2, wherein the arithmetic decod- ing is configured to determine (1110a-1110e) bit positions (numbits, numbits-1) of the at least two most significant bits and to allocate (1110f, 1110g) the at least two most significant bits determined by a symbol (sym) of the arithmetically encoded representation to the determined bit positions

4. The audio decoder according to one of claims 1 to 3, wherein the audio decoder is configured to decode (954; 1120a-1120e), for all spectral values for which two or more most significant bits have been decoded and which comprise more bits than the two or more most significant bits and a least significant bit, one or more inter- mediate bits, bit positions of which are between the least significant bit and the two or more most significant bits.

5. The audio decoder according to one of claims 1 to 4, wherein the audio decoder is configured to decode, in a first decoding phase,

- two or more most significant bits per spectral value (950; 1110a-1110g), and - for all spectral values for which two or more most significant bits are decoded and which comprise more bits than the two or more most significant bits and a least significant bit, one or more intermediate bits (954; 1120a-1120e), bit posi- tions of which are between the least significant bit and the two or more most significant bits, and

- for all spectral values for which two or more most significant bits are decoded and for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, signs (958; 1130a-1131d); and

wherein the audio decoder is configured to selectively omit (1130a, 1131a), in the first decoding phase, a decoding of a sign for spectral values for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and

wherein the audio decoder is configured to selectively obtain (972; 1140h- 1140j, 1141h-1141j), in a second decoding phase which follows the first decoding phase, sign information for spectral values for which the two or more most signifi- cant values and any intermediate bits, as far as intermediate bits are present, indi- cate a zero value and for which a least significant bit information indicates a non- zero value.

6. The audio decoder according to one of claims 1 to 5, wherein the audio decoder is configured to sequentially use (1140e, 1140i, 1141e, 1141i) subsequent bits of a least-significant-bit-information bit sequence (Isbs[]) in order to obtain least signifi- cant bit values associated with the spectral values.

7. The audio decoder according to claim 6, wherein the audio decoder is configured to use (1140e, 1141e) a single bit of the least-significant-bit-information bitse- quence for respective spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the used single bit of the least-significant-bit-information bitsequence is used (1140f, 1140g, 1141f, 1141g) in order to obtain a least signifi- cant bit value; and

wherein the audio decoder is configured to use (1140e, 1141e) a single bit of the least-significant-bit-information bitsequence for respective spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which the used single bit of the least-significant-bit-information bitsequence confirms the zero value; and

wherein the audio decoder is configured to use (1140e, 11401, 1141e, 1141i) two subsequent bits of the least-significant-bit-information bitsequence for respective spectral values for which the two or more most significant values and any inter- mediate bits, as far as intermediate bits are present, indicate a zero value, and for which a first of the used bits of the least-significant-bit-information bitsequence in- dicates a deviation from the zero value by a least significant bit value, wherein a second of the used bits of the least-significant-bit-information bitsequence deter- mines (1140j, 1141j) a sign of the respective spectral value.

8. The audio decoder according to one of claims 1 to 7, wherein the audio decoder is configured to decode (972; 1140a-1141j) least significant bits starting from a least significant bit associated with a lowest frequency spectral value and proceeding towards spectral values associated with increasingly higher frequencies,

such that spectral values are refined by least-significant-bit information in a range from a lowest frequency spectral value up to a spectral value for which a last least significant bit information is available, and such that spectral values having associ- ated frequencies higher than a frequency associated with the spectral value for which the last least significant bit information is available remain unrefined.

9. The audio decoder according to one of claims 1 to 8, wherein the audio decoder is configured to be switchable between

- a first mode (930,934,938,942,944,948) in which a decoding of spectral values in a higher frequency range is omitted in response to a signaling from the en- coder and in which least significant bits are decoded (934) for all spectral val- ues for which one or more most significant bits are decoded and which com- prise more bits than the most significant bits, and

- a second mode (950,954,958,962,968,972) in which one or more least signifi- cant bits associated with one or more of the spectral values are decoded (972), while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits are decoded and which comprise more bits than the most significant bits.

10. The audio decoder according to claim 9, wherein the audio decoder is configured to evaluate a bitstream flag which is included in the encoded audio information in order to decide whether the audio decoder operates in the first mode or in the sec- ond mode.

11. The audio decoder according to one of claims 1 to 10, wherein the audio decoder is configured to jointly decode (950; 1110a-1110g) two or more most significant bits per spectral value for at least two spectral values (Xq[n],Xq[n+1]) on the basis of respective symbol codes,

wherein a respective symbol code represents two or more most significant bits per spectral value for at least two spectral values.

12. The audio decoder according to one of claims 1 to 11, wherein the audio decoder is configured to decode spectral values according to the following algorithm:

Decode the 2 most significant bits of both coefficients Xq(n) and Xq(n+1) ac- cording to:

numbits = 1 ;

do {

Get probabilities p from context c

Decode symbol sym with arithmetic decoding and probabilities p

Update context c

numbits++;

} while (sym==VAL_ESC)

Xq[n] = (sym & 3) « (numbits-2);

Xq[n+1] = (sym » 2) « (numbits-2);

Decode remaining bits except the least significant bit, if there are any remaining bits, according to:

for (b = 1 ; b < numbits-2; b++) {

Decode bit0

Xq[n] += bit0 « b

Decode bit1

Xq[n+1] += bit1 « b

}

Decode the sign of each coefficient, except if the most significant bit is zero and the remaining bits are zero according to:

if(Xq[n] !=0 ) {

Decode bit0

if (bit0 == 1) {

Xq[n] = -Xq[n];

}

}

if(Xq[n+1] !=0) {

Decode bit1

if (bit1 == 1) {

Xq[n+1] = -Xq[n+1];

}

}

Set all coefficients n >= lastnz to zero, wherein lastnz is obtained on the basis of a side information obtained from the encoded audio representation;

Finalize the arithmetic decoding and compute the number of unused bits;

if there are unused bits, decode nlsbs bits and store them in a data structure lsbs [ ] ;

then refine the coefficients (Xq(n),Xq(n+1)) if numbits[n]>2 using the decoded LSB bits according to:

k = 0;

for (n = 0; n < lastnz; n+=2) {

if (numbits[n] > 2) {

if (k == nlsbs) {

break;

}

bit0 = lsbs[k++];

if (bit0 == 1) {

if (Xq[n] > 0) {

Xq[n] += 1 ;

} else if (Xq[n] < 0) {

Xq[n] -= 1 ;

} else {

if (k == nlsbs) {

break;

}

bit1 = lsbs[k++];

Xq[n] = 1 - 2*bit1 ;

}

}

if (k == nlsbs) {

break;

}

bit0 = lsbs[k++];

if (bit0 == 1 ) {

if (Xq[n+1] > 0) {

Xq[n+1]+= 1 ;

} else if (Xq[n+1 ] < 0) {

Xq[n+ 1 ] -= 1 ;

} else {

if (k == nlsbs) {

break;

}

bit1 = lsbs[k++];

Xq[n+1] = 1 - 2*bit1 ;

}

}

}

}

13. An audio decoder (100;200;700) for providing a decoded audio information (112;212;712) on the basis of an encoded audio information (110;210;710), wherein the audio decoder is configured to obtain decoded spectral values (132;232;732; Xq[n], Xq[n+1]) on the basis of an encoded information (130;230) representing the spectral values,

wherein the audio decoder is configured to decode (950; 1110a-1110g) one or more most significant bits on the basis of respective symbol codes (sym) for a plu- rality of spectral values (Xq[0]...Xq[lastnz-1]), and to decode one or more least sig- nificant bits for one or more of the spectral values,

wherein the audio decoder is configured to be switchable between

- a first mode (930,934,938,942,944,948) in which a decoding of spectral values in a higher frequency range is omitted in response to a signaling from the en- coder and in which least significant bits are decoded (934) for all spectral val- ues for which one or more most significant bits are decoded and which com- prise more bits than the most significant bits, and

- a second mode (950,954,958,962,968,972) in which one or more least signifi- cant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits have been decoded and which com- prise more bits than the most significant bits; and

wherein the audio decoder is configured to provide the decoded audio information using the spectral values.

14. The audio decoder according to claim 13, wherein the arithmetic decoding is con- figured to determine (950; 1110a-1110e) bit positions (numbits, numbits-1) of the one or more most significant bits and to allocate (1110f, 1110g) the one or more most significant bits determined by a symbol of the arithmetically encoded repre- sentation to the determined bit positions

15. The audio decoder according to one of claims 13 to 14, wherein the audio decoder is configured to decode (954; 1120a-1120e), for all spectral values for which one or more most significant bits have been decoded and which comprise more bits than the one or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits.

16. The audio decoder according to one of claims 13 to 15, wherein the audio decoder is configured to decode, when operating in the second mode, in a first decoding phase,

- one or more most significant bits per spectral value, and

- for all spectral values for which one or more most significant bits are decoded and which comprise more bits than the one or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are be- tween the least significant bit and the one or more most significant bits, and - for all spectral values for which one or more most significant bits are decoded and for which the one or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, signs; and

wherein the audio decoder is configured to selectively omit, when operating in the second mode, in the first decoding phase, a decoding of a sign for spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and

wherein the audio decoder is configured to selectively obtain, when operating in the second mode, in a second decoding phase which follows the first decoding phase, sign information for spectral values for which the one or more most signifi- cant values and any intermediate bits, as far as intermediate bits are present, indi- cate a zero value and for which a least significant bit information indicates a non- zero value.

17. The audio decoder according to one of claims 13 to 16, wherein the audio decoder is configured to sequentially use subsequent bits of a least-significant-bit- information bit sequence in order to obtain least significant bit values associated with the spectral values when operating in the second mode.

18. The audio decoder according to claim 17, wherein the audio decoder is configured to use, when operating in the second mode, a single bit of the least-significant-bit- information bitsequence for respective spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the used single bit of the least- significant-bit-information bitsequence is used in order to obtain a least significant bit value; and

wherein the audio decoder is configured to use, when operating in the second mode, a single bit of the least-significant-bit-information bitsequence for respective spectral values for which the one or more most significant values and any inter- mediate bits, as far as intermediate bits are present, indicate a zero value, and for which the used single bit of the least-significant-bit-information bitsequence con- firms the zero value; and

wherein the audio decoder is configured to use, when operating in the second mode, two subsequent bits of the least-significant-bit-information bitsequence for respective spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which a first of the used bits of the least-significant-bit-information bitse- quence indicates a deviation from the zero value by a least significant bit value, wherein a second of the used bits of the least-significant-bit-information bitse- quence determines a sign of the respective spectral value.

19. The audio decoder according to one of claims 13 to 18, wherein the audio decoder is configured to decode, when operating in the second mode, least significant bits starting from a least significant bit associated with a lowest frequency spectral val- ue and proceeding towards spectral values associated with increasingly higher frequencies,

such that spectral values are refined by least-significant-bit information in a range from a lowest frequency spectral value up to a spectral value for which a last least significant bit information is available, and such that spectral values having associ- ated frequencies higher than a frequency associated with the spectral value for which the last least significant bit information is available remain unrefined.

20. The audio decoder according to one of claims 13 to 19, wherein the audio decoder is configured to evaluate a bitstream flag which is included in the encoded audio information in order to decide whether the audio decoder operates in the first mode or in the second mode.

21. The audio decoder according to one of claims 13 to 20, wherein the audio decoder is configured to obtain (934) intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and the least sig- nificant bit associated with a given spectral value from a contiguous bit sequence in the first mode, and

wherein the audio decoder is configured to obtain (954) intermediate bits, bit posi- tions of which are between the least significant bit and the one or more most signif- icant bits, and the least significant bit associated with a given spectral value (972) from separate bit sequences or from separate, non-contiguous bit locations of a bit sequence in the second mode.

22. The audio decoder according to one of claims 13 to 21, wherein the audio decode is configured to selectively obtain (938) a sign information associated with a spec- tral value only after a decoding of the one or or most significant bits, any interme- diate bits, bit positions of which are between the least significant bit and the one o more most significant bits, and the least significant bit associated with a given spectral value in the first mode, in dependence on whether the one or more most significant bits, the intermediate bits and the least significant bit indicate a zero value or not, and

wherein the audio decoder is configured to selectively obtain (958; 1130a-1131d) a sign information associated with a spectral value after a decoding of the one or ore most significant bits and any intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, but before a least significant bit associated with a given spectral value is decoded in the second mode, in dependence on whether the one or more most significant bits and the in- termediate bits indicate a zero value or not.

23. An audio encoder (300;400;500;600) for providing an encoded audio information (312;412;512;612) on the basis of an input audio information (310;410;510;610), wherein the audio encoder is configured to obtain (620,630,640,650,660) spectral values (330;662;Xq[n]) representing an audio content of the input audio infor- mation, and

wherein the audio encoder is configured to encode (670;800) at least a plurality of the spectral values, in order to obtain an encoded information (350,450,550,672; sym,lsbs[]) representing the spectral values;

wherein the audio encoder is configured to jointly encode (878,886,890;

1000a, 1020a, 1040a-1040d) two or more most significant bits per spectral value, to obtain respective symbol codes (sym) for a set of spectral values (Xq[0]...Xq[lastnz-1]) using an arithmetic encoding,

wherein a respective symbol code (sym) represents two or more most significant bits per spectral value for one or more spectral values,

wherein the audio encoder is configured to encode (882;898; 1010a-1010e, 1011a- 1011e) one or more least significant bits associated with one or more of the spec- tral values in dependence on a bit budget available,

such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits; and

wherein the audio encoder is configured to provide the encoded audio information using the encoded information representing the spectral values.

24. The audio encoder according to claim 23, wherein the arithmetic encoding is con- figured to determine (878; 1000a) bit positions (numbits, numbits-1) of the at least two most significant bits and include (886,1020a) into the arithmetically encoded representation an information describing the bit positions.

25. The audio encoder according to claim 23 or claim 24, wherein the audio encoder is configured to map (890; 1040a-1040d) at least two most significant bits of the at least one spectral value (Xq[n,Xq[n+1) onto one symbol (sym) of an arithmetically encoded representation, which represents the at least two most significant bits of the at least one spectral value.

26. The audio encoder according to one of claims 23 to 25, wherein the audio encoder is configured to encode (892; 1050a-1050c), for all spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are between the least significant bit and the two or more most significant bits.

27. The audio encoder according to one of claims 23 to 26, wherein the audio encoder is configured to encode, in a first encoding phase,

- two or more most significant bits per spectral value

(878,886,890; 1000a, 1020a, 1040a-1040d), and

- for all spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are be- tween the least significant bit and the two or more most significant bits (892; 1050a-1050c), and

- for all spectral values for which two or more most significant bits are encoded and for which the two or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, signs (894;1060a-1061c); and

wherein the audio encoder is configured to selectively omit (1060a, 1061a), in the first encoding phase, an encoding of a sign for spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and

wherein the audio encoder is configured to selectively encode (898), in a second encoding phase, sign information (1010e, 1011e) for spectral values for which the two or more most significant values and any intermediate bits, as far as intermedi- ate bits are present, indicate a zero value and for which a least significant bit in- formation indicates a non-zero value.

28. The audio encoder according to one of claims 23 to 27, wherein the audio encoder is configured to only include (882,898; 1010c-1010e, 1011c-1011e) a sign infor- mation into the encoded audio representation for spectral values which only differ from zero by a least significant bit if the least significant bit of such spectral values is actually encoded.

29. The audio encoder according to one of claims 23 to 28, wherein the audio encoder is configured to sequentially provide subsequent bits of a least-significant-bit- information bit sequence (Isbs[]) in order to encode least significant bit values as- sociated with the spectral values.

30. The audio encoder according to claim 29, wherein the audio encoder is configured to provide (882,898; 1010a, 1010b, 1011a,1011b) a single bit (bit) of the least- significant-bit-information bitsequence (lsbs[]) for respective spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the used single bit of the least-significant-bit-information bitsequence is used in order to encode a least significant bit value; and

wherein the audio encoder is configured to provide

(882, 898; 1010a, 1010b, 1011a, 1011b) a single bit of the least-significant-bit- information bitsequence for respective spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which the provided single bit (bit) of the least-significant-bit-information bitsequence confirms the zero value; and wherein the audio encoder is configured to provide (882,898;

1010a, 1010b, 1010d, 1010e, 1011a, 1011b, 1011d, 1011e) two subsequent bits of the least-significant-bit-information bitsequence for respective spectral values for which the two or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which a first of the pro- vided bits of the least-significant-bit-information bitsequence indicates a deviation from the zero value by a least significant bit value, wherein a second of the provid- ed bits of the least-significant-bit-information bitsequence encodes a sign of the re- spective spectral value.

31. The audio encoder according to one of claims 23 to 30, wherein the audio encoder is configured to encode (882,898; 1140a-1141j) least significant bits starting from a least significant bit associated with a lowest frequency spectral value and proceed- ing towards spectral values associated with increasingly higher frequencies,

such that encoded information for refining spectral values by least-significant-bit in- formation is provided in a range from a lowest frequency spectral value up to a spectral value for which a last least significant bit information is provided, and such that no encoded information for refining spectral values by least-significant-bit information is provided for spectral values having associated frequencies higher than a frequency associated with the spectral value for which the last least signifi- cant bit information is provided.

32. The audio encoder according to one of claims 23 to 31, wherein the audio encoder is configured to be switchable between

- a first mode (840,844,848,852,856,860,864,868,869) in which an encoding of non-zero spectral values in a higher frequency range is omitted in case that an available bit budget is used up by an encoding of spectral values in a lower frequency range and in which least significant bits are encoded (860) for all

spectral values for which one or more most significant bits are encoded (848,852,856) and which comprise more bits than the most significant bits, and

- a second mode (870,874,878,882,886,890,892,894,896,898) in which one or more least significant bits associated with one or more of the spectral values are encoded (898), while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits.

33. The audio encoder according to claim 32, wherein the audio encoder is configured to provide a bitstream flag which is included in the encoded audio information in order to indicate whether the audio encoder operates in the first mode or in the second mode.

34. The audio encoder according to one of claims 23 to 33, wherein the audio encoder is configured to jointly encode (878,886,890; 1000a, 1020a, 1040a-1040d) two or more most significant bits per spectral value for at least two spectral values (Xq[n],Xq[n+1]) using respective symbol codes (sym),

wherein a respective symbol code represents two or more most significant bits per spectral value for at least two spectral values.

35. The audio encoder according to one of claims 23 to 34, wherein the audio encoder is configured determine an actual highest-frequency non-zero spectral value and to encode at least two or more most significant bits of all non-zero spectral values or of all non-zero groups of spectral values.

36. The audio encoder according to one of claims 23 to claim 35, wherein the audio encoder is configured to encode (878,886,890,892) all bits except for a least signif- icant bit for all non-zero spectral values, and

wherein the audio encoder is configured to encode (882,898) least significant bits for spectral values until a bit budget is exhausted.

37. The audio encoder according to one of claims 23 to 36, wherein the audio encoder is configured to obtain (810; 814,818,82) a gain information which determines

quantization steps of a quantization (660) of spectral values, and which determines a bit demand for encoding the quantized spectral values.

38. An audio encoder (300;400;500;600) for providing an encoded audio information (312;412;512;612) on the basis of an input audio information (310;410;510;610), wherein the audio encoder is configured to obtain (620,630,640,650,660) spectral values representing an audio content of the input audio information, and wherein the audio encoder is configured to encode (670;800) at least a plurality of the spectral values, in order to obtain an encoded information (350,450,550,672; sym.lsbs[]) representing the spectral values;

wherein the audio encoder is configured to encode (882;898; 1010a-1010e, 1011a- 1011e) one or more most significant bits, to obtain respective symbol codes for a plurality of the spectral values (Xq[0]...Xq[lastnz-1]), and to encode one or more least significant bits for one or more of the spectral values,

wherein a respective symbol code (sym) represents one or more most significant bits values for one or more spectral values (Xq[n], Xq[n+1]),

wherein the audio encoder is configured to be switchable between

- a first mode (840,844,848,852,856,860,864,868,869) in which an encoding of non-zero spectral values in a higher frequency range is omitted in case that an available bit budget is used up by an encoding of spectral values in a lower frequency range and in which least significant bits are encoded (860) for all spectral values for which one or more most significant bits are encoded (848,852,856) and which comprise more bits than the most significant bits, and

- a second mode (870,874,878,882,886,890,892,894,896,898) in which one or more least significant bits associated with one or more of the spectral values are encoded (898), while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits;

wherein the audio encoder is configured to provide the encoded audio information using the encoded information representing the spectral values.

39. The audio encoder according to claim 38, wherein the audio encoder is configured to encode (878,886,890) at least one or more most significant bits of all non-zero spectral values or of all non-zero groups of spectral values in the second mode.

40. The audio encoder according to one of claims 38 to 39, wherein the audio encoder is configured to limit, when operating in the first mode, a frequency range, for which spectral values are encoded, in case that a bit budget is insufficient, such that one or more spectral values are left unconsidered in the encoding of spectral values.

41. The audio encoder according to claim 40, wherein the audio encoder is configured to determine, when operating in the first mode, a maximum frequency value and to encode (848,852,856,860), when operating in the first mode, spectral values up to the maximum frequency and to leave, when operating in the first mode, spectral values above the maximum frequency unencoded even if the spectral values are non-zero,

wherein the audio encoder is configured to select, when operating in the first mode, the maximum frequency value in dependence on a computation or estima- tion of a bit demand for encoding all spectral values, such that a number of spec- tral values to be encoded is reduced if the computed or estimated bit demand would exceed a bit budget, and

wherein the audio encoder is configured to determine, when operating in the sec- ond mode, the maximum frequency value and to encode (878,882,886,890,898) when operating in the second mode, spectral values up to the maximum frequency and to leave, when operating in the second mode, spectral values above the max- imum frequency unencoded,

wherein the audio encoder is configured to select, when operating in the second mode, the maximum frequency value such that at least one or more most signifi- cant bits of all non-zero spectral values or of all non-zero groups of spectral values are encoded and such that at most zero-valued spectral values are left unencoded.

42. The audio encoder according to claim 40 or claim 41, wherein the audio encoder is configured to include an information (lastnz) describing the maximum frequency in- to the encoded audio information.

43. The audio encoder according to one of claims 38 to 42, wherein the audio encoder is configured to make a mode decision (830) whether to use the first mode or the second mode in dependence on an available bit rate.

44. The audio encoder according to one of claims 38 to 43, wherein the audio encoder is configured to make a mode decision (830) whether to use the first mode or the second mode in dependence on a number of spectral values or groups of spectral values which comprise, in addition to one or more most significant bits encoded in a most-significant-bit-encoding step, one or more least significant bits, an encoding of which can selectively be omitted in dependence on a bit demand and a bit budget.

45. The audio encoder according to one of claims 38 to 44, wherein the audio encoder is configured to include a bitstream flag in the encoded audio information indicating whether the audio encoder operates in the first mode or in the second mode.

46. The audio encoder according to one of claims 38 to 45, wherein the audio encoder is configured to encode (860) intermediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and the least sig- nificant bit associated with a given spectral value into a contiguous bit sequence in the first mode, and

wherein the audio encoder is configured to encode (892) intermediate bits, bit posi- tions of which are between the least significant bit and the one or more most signif- icant bits, and the least significant bit associated with a given spectral value (882,898) into separate bit sequences or into separate, non-contiguous bit loca- tions of a bit sequence in the second mode.

47. The audio encoder according to one of claims 38 to 46, wherein the audio encoder is configured to encode (864), when operating in the first mode, a sign information associated with a spectral value in a bit sequence which is associated with inter- mediate bits, bit positions of which are between the least significant bit and the one or more most significant bits, and least significant bits, and

wherein the audio encoder is configured to selectively encode, when operating in the second mode, a sign information associated with a spectral value in a bit se- quence which is associated with intermediate bits, bit positions of which are be- tween the least significant bit and the one or more most significant bits, and sign information (894) or in a bit sequence (Isbs[]) associated with least significant bits and sign information (882,898), such that sign information for spectral values which deviate from zero only by a least significant bit value are encoded in the bit sequence associated with least significant bits and sign information.

48. An audio encoder (300;400;500;600) for providing an encoded audio information (312;412;512;612) on the basis of an input audio information(310;410;510;610), wherein the audio encoder is configured to obtain (620,630,640,650,660) spectral values (330;662;Xq[n]) representing an audio content of the input audio infor- mation, and

wherein the audio encoder is configured to encode (670;800) at least a plurality of the spectral values, in order to obtain an encoded information (350,450,550,672; sym.Isbs[]) representing the spectral values;

wherein the audio encoder is configured to obtain (810,814,818,822) a gain infor- mation which determines quantization steps of a quantization of spectral values, and which determines a bit demand for encoding the quantized spectral values (330;662;Xq[n]);

wherein the audio encoder is configured to encode (878,886,890;

1000a, 1020a, 1040a-1040d) one or more most significant bits using respective symbol codes (sym) for a plurality of the spectral values (Xq[0]...Xq[lastnz-1]) us- ing an arithmetic encoding, and to encode one or more least significant bits for one or more of the spectral values,

wherein a respective symbol code (sym) represents one or more most significant bits per spectral value for one or more spectral values,

wherein the audio encoder is configured to encode (882;898;1010a-1010e, 1011a- 1011e) one or more least significant bits associated with one or more of the spec- tral values in dependence on a bit budget available,

such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the one or more most significant bits ; and

wherein the audio encoder is configured to provide the encoded audio information using the encoded information representing the spectral values.

49. The audio encoder of claim 48, wherein the audio encoder is configured to obtain (810) a first estimate of the gain information based on an energy of groups of spectral values,

to quantize (814) set of spectral values using the first estimate of gain information,

to compute or estimate (818) a number of bits needed to encode the set of spectral values quantized using the first estimate of gain information or using a refined gain information, and

to decide (830) whether to use the first mode or the second mode in dependence on a number of bits needed.

50. The audio encoder according to claim 48 or claim 49,

wherein the audio encoder is configured to be switchable between

- a first mode in which an encoding of non-zero spectral values in a higher fre- quency range is omitted in case that an available bit budget is used up by en- coded spectral values in a lower frequency range and in which least significant bits are encoded for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and

- a second mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most sig- nificant bits are encoded and which comprise more bits than the most signifi- cant bits; and

wherein the audio encoder is configured to decide whether to use the first mode or the second mode in dependence on the number of bits needed and in dependence

on a criterion which indicates how many spectral values comprise more bits than the one or more most significant bits.

51. The audio encoder according to claim 48 or claim 49,

wherein the audio encoder is configured to be switchable between

- a first mode in which an encoding of non-zero spectral values in a higher fre- quency range is omitted in case that an available bit budget is used up by en- coded spectral values in a lower frequency range and in which least significant bits are encoded for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and

- a second mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most sig- nificant bits are encoded and which comprise more bits than the most signifi- cant bits; and

wherein the audio encoder is configured to decide whether to use the first mode or the second mode in dependence on the number of bits needed and in dependence on a bitrate, such that the second mode is chosen if a bitrate is larger than or equal to a threshold bitrate and if a computed or estimated number of bits needed to en- code the set of spectral values is higher than a bit budget.

52. The audio encoder according to one of claims 38 to 51, wherein the arithmetic en- coding is configured to determine bit positions of the one or more most significant bits and include into the arithmetically encoded representation an information de- scribing the bit positions.

53. The audio encoder according to one of claims 38 to 52, wherein the audio encoder is configured to map at least two most significant bits of the at least one spectral value onto one symbol of an arithmetically encoded representation, which repre- sents the at least two most significant bits of the at least one spectral value.

54. The audio encoder according to one of claims 38 to 53, wherein the audio encoder is configured to encode, for all spectral values for which one or more most signifi- cant bits are encoded and which comprise more bits than the one or more most significant bits and a least significant bit, one or more intermediate bits, bit posi- tions of which are between the least significant bit and the one or more most signif- icant bits.

55. The audio encoder according to one of claims 38 to 54, wherein the audio encoder is configured to encode, in a first encoding phase,

- one or more most significant bits per spectral value, and

- for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the one or more most significant bits and a least significant bit, one or more intermediate bits, bit positions of which are be- tween the least significant bit and the one or more most significant bits, and - for all spectral values for which one or more most significant bits are encoded and for which the one or more most significant bits and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, signs; and

wherein the audio encoder is configured to selectively omit, in the first encoding phase, an encoding of a sign for spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are pre- sent, indicate a zero value, and

wherein the audio encoder is configured to selectively encode, in a second encod- ing phase which follows the first encoding phase, sign information for spectral val- ues for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value and for which a least significant bit information indicates a non-zero value.

56. The audio encoder according to one of claims 38 to 55, wherein the audio encoder is configured to only include a sign information into the encoded audio representa- tion for spectral values which only differ from zero by a least significant bit if the least significant bit of such spectral values is actually encoded.

57. The audio encoder according to one of claims 38 to 56, wherein the audio encoder is configured to sequentially provide subsequent bits of a least-significant-bit- information bit sequence in order to encode least significant bit values associated with the spectral values.

58. The audio encoder according to claim 57, wherein the audio encoder is configured to provide a single bit of the least-significant-bit-information bitsequence for re- spective spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a non-zero value, wherein the used single bit of the least-significant-bit-information bitsequence is used in order to encode a least significant bit value; and

wherein the audio encoder is configured to provide a single bit of the least- significant-bit-information bitsequence for respective spectral values for which the one or more most significant values and any intermediate bits, as far as intermedi- ate bits are present, indicate a zero value, and for which the provided single bit of the least-significant-bit-information bitsequence confirms the zero value; and wherein the audio encoder is configured to provide two subsequent bits of the least-significant-bit-information bitsequence for respective spectral values for which the one or more most significant values and any intermediate bits, as far as intermediate bits are present, indicate a zero value, and for which a first of the pro- vided bits of the least-significant-bit-information bitsequence indicates a deviation from the zero value by a least significant bit value, wherein a second of the provid- ed bits of the least-significant-bit-information bitsequence encodes a sign of the re- spective spectral value.

59. The audio encoder according to one of claims 38 to 58, wherein the audio encoder is configured to encode least significant bits starting from a least significant bit as- sociated with a lowest frequency spectral value and proceeding towards spectral values associated with increasingly higher frequencies,

such that encoded information for refining spectral values by least-significant-bit in- formation is provided in a range from a lowest frequency spectral value up to a spectral value for which a last least significant bit information is provided, and such that no encoded information for refining spectral values by least-significant-bit information is provided for spectral values having associated frequencies higher than a frequency associated with the spectral value for which the last least signifi- cant bit information is provided.

60. The audio encoder according to one of claims 38 to 59, wherein the audio encoder is configured to be switchable between

- a first mode in which an encoding of non-zero spectral values in a higher fre- quency range is omitted in case that an available bit budget is used up by en- coded spectral values in a lower frequency range and in which least significant bits are encoded for all spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits, and

- a second mode in which one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most sig- nificant bits are encoded and which comprise more bits than the most signifi- cant bit.

61. The audio encoder according to claim 60, wherein the audio encoder is configured to provide a bitstream flag which is included in the encoded audio information in order to indicate whether the audio encoder operates in the first mode or in the second mode.

62. The audio encoder according to one of claims 38 to 61 , wherein the audio encoder is configured to jointly encode one or more most significant bits per spectral value for at least two spectral values using respective symbol codes,

wherein a respective symbol code represents one or more most significant bits per spectral value for at least two spectral values.

63. The audio encoder according to one of claims 38 to 62, wherein the audio encoder is configured to determine, when operating in the second mode, an actual highest- frequency non-zero spectral value and to encode at least one or more most signifi- cant bits of all non-zero spectral values or of all non-zero groups of spectral val- ues.

64. The audio encoder according to one of claims 38 to 63, wherein the audio encoder is configured to encode, when operating in the second mode, all bits except for a least significant bit for all non-zero spectral values, and

wherein the audio encoder is configured to encode least significant bits for spectral values until a bit budget is exhausted.

65. The audio encoder according to one of claims 38 to 64, wherein the audio encoder is configured to obtain a gain information which determines quantization steps of a quantization of spectral values, and which determines a bit demand for encoding the quantized spectral values.

66. A method for providing a decoded audio information (112; 212; 712) on the basis of an encoded audio information (110;210;710),

wherein the method comprises obtaining decoded spectral values (132;232;732; Xq[n], Xq[n+1]) on the basis of an encoded information (130; 230) representing the spectral values,

wherein the method comprises jointly decoding (950; 1110a-1110g) two or more most significant bits per spectral value on the basis of respective symbol codes (sym) for a set of spectral values using an arithmetic decoding,

wherein a respective symbol code (sym) represents two or more most significant bits per spectral value for one or more spectral values,

wherein the method comprises decoding (972; 1140a-1141j) one or more least significant bits associated with one or more of the spectral values in dependence on how much least significant bit information is available,

such that one or more least significant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits are decoded and which comprise more bits than the one or more most significant bits; and

wherein the method comprises providing the decoded audio information using the spectral values.

67. A method for providing a decoded audio information (112;212;712) on the basis of an encoded audio information (110;210;710),

wherein the method comprises obtaining decoded spectral values (132;232;732; Xq[n], Xq[n+1]) on the basis of an encoded information (130;230) representing the spectral values,

wherein the method comprises decoding (950; 1110a-1110g) one or more most significant bits on the basis of respective symbol codes (sym) for a plurality of spectral values (Xq[0]...Xq[lastnz-1]), and

decoding one or more least significant bits for one or more of the spectral values,

wherein the method comprises selecting between

- a first mode (930,934,938,942,944,948) in which a decoding of spectral values in a higher frequency range is omitted in response to a signaling from the en- coder and in which least significant bits are decoded (934) for all spectral val- ues for which one or more most significant bits are decoded and which com- prise more bits than the most significant bits, and

- a second mode (950,954,958,962,968,972) in which one or more least signifi- cant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits have been decoded and which com- prise more bits than the most significant bits; and

wherein the method comprises providing the decoded audio information using the spectral values.

68. A method for providing an encoded audio information (312;412;512;612) on the basis of an input audio information (310;410;510;610),

wherein the method comprises obtaining (620,630,640,650,660) spectral values (330;662;Xq[n]) representing an audio content of the input audio information, and wherein the method comprises encoding (670;800) at least a plurality of the spec- tral values, in order to obtain an encoded information (350,450,550,672; sym,Isbs[]) representing the spectral values;

wherein the method comprises jointly encoding (878,886,890;

1000a, 1020a, 1040a-1040d) two or more most significant bits per spectral value, to obtain respective symbol codes (sym) for a set of spectral values (Xq[0]...Xq[lastnz-1]), using an arithmetic encoding,

wherein a respective symbol code (sym) represents two or more most significant bits per spectral value for one or more spectral values,

wherein the method comprises encoding (882;898; 1010a-1010e, 1011a-1011e) one or more least significant bits associated with one or more of the spectral val- ues in dependence on a bit budget available,

such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which two or more most significant bits are encoded and which comprise more bits than the two or more most significant bits; and

wherein the method comprises providing the encoded audio information using the encoded information representing the spectral values.

69. A method for providing an encoded audio information (312;412;512;612) on the basis of an input audio information (310;410;510;610),

wherein the method comprises obtaining (620,630,640,650,660) spectral values representing an audio content of the input audio information, and

wherein the method comprises encoding (670;800) at least a plurality of the spec- tral values, in order to obtain an encoded information (350,450,550,672; sym,Isbs[]) representing the spectral values;

wherein the method comprises encoding (882;898;1010a-1010e, 1011a-1011e) one or more most significant bits, to obtain respective symbol codes for a plurality of the spectral values (Xq[0]...Xq[lastnz-1]), and encoding one or more least signif- icant bits for one or more of the spectral values,

wherein a respective symbol code (sym) represents one or more most significant bits values for one or more spectral values (Xq[n], Xq[n+1]),

wherein the method comprises selecting between

- a first mode (840,844,848,852,856,860,864,868,869) in which an encoding of non-zero spectral values in a higher frequency range is omitted in case that an available bit budget is used up by an encoding of spectral values in a lower frequency range and in which least significant bits are encoded (860) for all spectral values for which one or more most significant bits are encoded (848,852,856) and which comprise more bits than the most significant bits, and

- a second mode (870,874,878,882,886,890,892,894,896,898) in which one or more least significant bits associated with one or more of the spectral values are encoded (898), while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the most significant bits;

wherein the method comprises providing the encoded audio information using the encoded information representing the spectral values.

70. A method for providing an encoded audio information (312;412;512;612) on the basis of an input audio information (310;410;510;610),

wherein the method comprises obtaining (620,630,640,650,660) spectral values (330; 662; Xq[n]) representing an audio content of the input audio information, and wherein the method comprises encoding (670;800) at least a plurality of the spec- tral values, in order to obtain an encoded information (350,450,550,672; sym, Isbs[]) representing the spectral values;

wherein the method comprises obtaining (810,814,818,822) a gain information which determines quantization steps of a quantization of spectral values, and which determines a bit demand for encoding the quantized spectral values;

wherein the method comprises encoding (878,886,890; 1000a, 1020a, 1040a- 1040d) one or more most significant bits using respective symbol codes (sym) for a plurality of the spectral values (Xq[0]...Xq[lastnz-1]) using an arithmetic encod- ing, and encoding one or more least significant bits for one or more of the spectral values,

wherein a respective symbol code (sym) represents one or more most significant bits per spectral value for one or more spectral values,

wherein the method comprises encoding (882;898; 1010a-1010e, 1011a-1011e) one or more least significant bits associated with one or more of the spectral val- ues in dependence on a bit budget available,

such that one or more least significant bits associated with one or more of the spectral values are encoded, while no least significant bits are encoded for one or more other spectral values for which one or more most significant bits are encoded and which comprise more bits than the one or more most significant bits ; and

wherein the method comprises providing the encoded audio information using the encoded information representing the spectral values.

71. A computer program for performing the method according to one of claims 66 to 70 when the computer program runs on a computer.

72. An encoded audio representation, comprising:

an encoded information (130;230) representing spectral values; and

a flag indicating whether an audio decoder should work

- in a first mode (930,934,938,942,944,948) in which a decoding of spectral val- ues in a higher frequency range is omitted in response to a signaling from the encoder and in which least significant bits are decoded (934) for all spectral values for which one or more most significant bits are decoded and which comprise more bits than the most significant bits, or

- in a second mode (950,954,958,962,968,972) in which one or more least sig- nificant bits associated with one or more of the spectral values are decoded, while no least significant bits are decoded for one or more other spectral values for which one or more most significant bits have been decoded and which comprise more bits than the most significant bits.