FIELD OF THE INVENTION
The present invention is in the field of coding, where different characteristics of data to be
encoded are utilized for coding rates, as for example in video and audio coding.
BACKGROUND OF THE INVENTION
State of the art coding strategies can make use of characteristics of a data stream to be encoded.
For example, in audio coding, perception models are used in order to compress source data
almost without decreasing the noticeable quality and degradation when replayed Modern
perceptual audio coding schemes, such as for example, MPEG-
2/4 AAC (MPEG = Moving Pictures Expert Group, AAC = Advanced Audio Coding), cf
Generic Coding of Moving Pictures and Associated Audio. Advanced Audio Coding,
International Standard 13818-7, ISO/IEC JTC1/SC29/WG11 Moving Pictures Expert Group,
1997, may use filter banks, such as for example the Modified Discrete Cosine Transform
(MDCT), for representing the audio signal in the frequency domain
In the frequency domain quantization of frequency coefficients can be carried out, according to a
perceptual model Such coders can provide excellent perceptual audio
quality for general types of audio signals as, for example,music. On the other hand, modern
speech. coders, such as, for example, ACELP (ACELP = Algebraic Code Excited Linear
Prediction), use a predictive approach, and in this way may represent the audio/speech signal in
the time domain Such speech coders can model the characteristics of the human speech
production process, i.e the human vocal tract and, consequently, achieve excellent performance
for speech signals at low bit rates. Conversely, perceptional audio coders do not achieve the level
of performance offered by

speech coders for speech signals coded at low bit rates, and using speech coders to represent
general, audio signals/music results in significant quality impairments
Conventional concepts provide a layered combination in which always all partial coders are
active, 1 e time-domain and frequency-domain encoders, and the final output
signal is calculated by combining the contributions of the partial coders for a given processed
time frame. A popular example of layered coding are MPEG-4 scalable speech/audio coding with
a speech coder as the base layer and a filterbank-based enhancement layer, cf. Bernhard Grill,
Karlheinz Brandenburg, "A Two-or Three-Stage Bit-Rate Scalable Audio Coding System",
Preprint Number 4132, 99* Convention of the AES (September 1995)
Conventional frequency-domain encoders can make use of MDCT filterbanks. The MDCT has
become a dominant filterbank for conventional perceptual audio coders because of its
advantageous properties. For example, it can provide a smooth cross-fade between processing
blocks Even if a signal in each processing block is altered differently, for example due to
quantization of spectral coefficients, no blocking artifacts due to abrupt transitions from block to
block occur because of the windowed overlap/add operations The MDCT uses the concept of
time-domain aliasing cancellation (TDAC)
The MDCT is a Fourier-related transform based on the type-IV discrete cosine transform, with
the additional property of being lapped It is designed to be performed in consecutive blocks of a
larger data set, where subsequent blocks are overlapped so that the last half of one block
coincides with the first half of the next block. This overlapping, in addition to an energy-
compaction quality of the DCT, makes the MDCT especially attractive for signal compression
applications, since it helps to avoid said artifacts stemming from the block boundaries As a
lapped

transform, the MDCT is a bit unusual compared to other Fourier-related transforms in that it has
half as many outputs as inputs, instead of the same number In particular, 2N real numbers are
transformed into N real numbers, where N is a positive integer
The inverse MDCT is also known as IMDCT Because there are different numbers of inputs and
outputs, at first glance it might seem that the MDCT should not be invertible. However, perfect
invertibility is achieved by adding the overlap rDMCTs of subsequent overlapping blocks,
causing the errors to cancel and the original data to be retrieved, i.e.
achieving TDAC
Therewith, the number of spectral values at the output of a filterbank is equal to the number of
time-domain input values at its input which is also referred to as critical
sampling
An MDCT filterbank provides a high-frequency selectivity and enables a high coding gain. The
properties of overlapping of blocks and critical sampling can be achieved
by utilizing the technique of time-domain aliasing cancellation, cf J Princen, A Bradley,
"Analysis/Synthesis Filter Bank Design Based on Time Domain Aliasing Cancellation", IEEE
Trans ASSP, ASSP-34(5) 1153-1161, 1986. Fig. 4 illustrates these effects of an MDCT
Fig 4 shows an MDCT input signal, in terms of an impulse along a time axis 400 at the top The
input signal 400 is then transformed by two consecutive windowing and MDCT
blocks, where the windows 410 are illustrated underneath the input signal 400 in Fig 4 The back
transformed individual windowed signals are displayed in Fig. 4 by the time lines 420 and 425
After the inverse MDCT, the first block produces an aliasing component with positive sign 420,
the second block produces an aliasing component with the same magnitude and

a negative sign 425 The aliasing components cancel each other after addition of the two output
signals 420 and 425 as shown in the final output 430 at the bottom of Fig. 4
In "Extended Adaptive Multi-Rate - Wideband (AMR-WB+)codec", 3GPP TS 26.290V6 3 0,
2005-06, Technical Specification the AMR-WB+ (AMR-WB = Adaptive Multi-Rate Wideband)
codec is specified According to section 5 2, the encoding algorithm at the core of the AMR-WB+
codec is based on a hybrid ACELP/TCX (TCX = Transform coded Excitation) model. For every
block of an input signal the encoder decides, either in an open loop or a closed loop mode which
encoding model, I e. ACELP or TCX, is best. The ACELP model is a time-domain, predictive
encoder, best suited for speech and transient signals The AMR-WB encoder is used in ACELP
modes Alternatively, the TCX model is a transform based encoder, and is more appropriate for
typical music samples
Specifically, the AMR-WB+ uses a discrete Fourier transform (DFT) for the transform coding
mode TCX. In order to allow a smooth transition between adjacent blocks, a windowing and
overlap is used. This windowing and overlap is necessary both for transitions between different
coding modes (TCX/ACELP) and for consecutive TCX frames Thus, the DFT together with the
windowing and overlap represents a filterbank that is not critically sampled. The filterbank
produces more frequency values than the number of new input samples, cf Fig. 4 in 3GPP TS
26.290V6 3 0 (3GPP = Third Generation Partnership Project, TS = Technical Specification)
Each TCX frame utilizes an overlap of 1/8 of the frame length which equals the number of new
input samples Consequently, the corresponding length of the DFT is 9/8 of the frame length
Considering the non-critically sampled DFT filterbank in the TCX, i e the number of spectral
values at the output of the filterbank is larger than the number of time-domain

input values at its input, this frequency domain coding mode is different from audio codecs such
as AAC (AAC = Advanced Audio Coding) which utilizes an MDCT, a critically sampled lapped
transform
The Dolby E codec is described in Fielder, Louis D, Todd, Craig C, "The Design of a Video
Friendly Audio Coding System for Distributing Applications", Paper Number 17-008, The AES
17th International Conference High-Quality Audio Coding (August 1999) and Fielder, Louis D.,
Davidson, Grant A , "Audio Coding Tools for Digital Television Distribution", Preprint Number
5104, 108th Convention of the AES (January 2000) The Dolby E codec utilizes the MDCT
filterbank In the design of this coding, special focus was put on the possibility to perform editing
in the coding domain To achieve this, special alias-free windows are used At the boundaries of
these windows a smooth-cross fade or splicing of different signal portions is possible. In the
above-referenced documents it is, for example, outlined, cf section 3 of "The Design of a Video
Friendly Audio Coding System for Distribution Applications", that this would not be possible by
simply using the usual MDCT windows which introduce time-domain aliasing. However, it is
also described that the removal of aliasing comes at the cost of an increased number
of transform coefficients, indicating that the resulting filterbank does not have the property of
critical sampling anymore
EP1396844 describes a unified lossy and lossless audio compression scheme which combines
lossy and lossless audio compression within a same audio signal This approach employs mixed
lossless coding of a transition frame between lossy and lossless coding frames to produce
seamless transitions The mixed lossless coding performs a lapped transform and inverse lapped
transform to produce an appropriately windowed and folded pseudo-time domain frame, which
can then be losslessly coded The mixed lossless coding also can be applied for frames that
exhibit poor lossy compression performance

US2006247928 teaches method and system for operating audio encoders in parallel. According
to this invention, the time needed to encode an input audio stream is reduced by dividing the
stream into two or more overlapping segments of audio information blocks, applying an encoding
process to each segment to generate encoded segments in parallel, and appending the encoded
segments to form an encoded output signal. The encoding process is responsive to one or more
control parameters Some of the control parameters, which apply to a given block, are calculated
from audio information in one or more previous blocks The length of the overlap between
adjacent segments is chosen such that the differences between control parameter values and
corresponding reference values at the end of the overlap interval are small enough to avoid
producing audible artifacts in a signal that is obtained by decoding the encoded output signal
US6226608 discloses an audio encoder applies an adaptive block-encoding process to segments
of audio information to generate frames of encoded information that are aligned with a reference
signal conveying the alignment of a sequence of video information frames The audio information
is analyzed to determine various characteristics of the audio signal such as the occurrence and
location of a transient, and a control signal is generated that causes the adaptive block-encoding
process to encode segments of varying length. A complementary decoder applies an adaptive
block-decoding process to recover the segments of audio information from the frames of encoded
information In embodiments that apply time-domain aliasing cancellation transforms, window
functions and transforms are applied according to one of a plurality of segment patterns that

The segments in each frame of a sequence of overlapping frames may be recovered without
aliasing artifacts independently from the recovery of segments in other frames Window functions
are adapted to provide preferred frequency-domain responses and time-domain gain profiles

US2005071402 discloses method of making a window type decision based on MDCT data in
audio encoding Accordingly, preliminary Modified Discrete Cosine Transform (MDCT)
coefficients are computed for a current frame of data and a next frame of data using a long
window type The computed preliminary MDCT coefficients of the current and next frames are
then used to determine the window type of the current frame. If the determined window type is
not the long window type, final MDCT coefficients are computed for the current frame using the
determined window type
OBJECTS OF THE INVENTION
It is the object of the present invention to provide a more efficient concept for encoding and
decoding data segments.

SUMMARY OF THE INVENTION
The present invention is based on the finding that a more efficient encoding and decoding concept
can be utilized by using combined time-domain and frequency-domain encoders,
respectively decoders. The problem of time aliasing can be efficiently combat by transforming
time-domain data to the frequency-domain in the decoder and by combining the resulting
transformed frequency-domain data with the decoded frequency-domain data received
Overheads can be reduced by adapting overlapping regions of overlap windows being applied to
data segments to coding domain changes. Using windows with smaller overlapping regions can
be beneficial when using time-domain encoding, respectively when switching from or to time-
domain encoding
Embodiments can provide a universal audio encoding and decoding concept that achieves
improved performance for both types of input signals, such as speech signals- and
music signals Embodiments can take advantage by combining multiple coding approaches, e g
time-domain and frequency- domain coding concepts Embodiments can efficiently combine
filterbank based and time-domain based coding concepts into a single scheme. Embodiments may
result in a combined codec which can, for example, be able to switch between an audio codec for
music-like audio content and a speech codec for speech-like content. Embodiments may utilize
this switching frequently, especially for mixed content
Embodiments of the present invention may provide the advantage that no switching artifacts
occur In embodiments the amount of additional transmit data, or additionally
coded samples, for a switching process can be minimized in order to avoid a reduced efficiency
during this phase of operation Therewith the concept of switched combination of partial coders is
different from that of the layered combination in which always all partial coders are active.

BRIEF DESCRIPTION OF THE ACCOMAPNYING DRAWINGS
In the following embodiments of the present invention will be described in detail using the
accompanying Figures, in which
Fig la shows an embodiment of an apparatus for decoding,
Fig lb shows another embodiment of an apparatus for decoding,
Fig. lc shows another embodiment of an apparatus for decoding;
Fig Id shows another embodiment of an apparatus for decoding,
Fig 1 e shows another embodiment of an apparatus for decoding,
Fig 1 f shows another embodiment of an apparatus for decoding,
Fig 2a shows an embodiment of an apparatus for encoding;
Fig. 2b shows another embodiment of an apparatus for encoding,
Fig 2c shows another embodiment of an apparatus for encoding,
Fig 3a illustrates overlapping regions when switching between frequency-domain and time-
domain coding for the duration of one window;
Fig 3b illustrates the overlapping regions when switching between frequency-domain coding and
time-domain coding for a duration of two windows,

Fig. 3c illustrates multiple windows with different overlapping regions,
Fig. 3d illustrates the utilization of windows with different overlapping regions in an
embodiment; and
Fig 4 illustrates time-domain aliasing cancellation when using MDCT
DETAIL DESCRIPTION OF THE PREFERRED EMBODIMENTS
Fig. la shows an apparatus 100 for decoding data segments representing a time-domain data
stream, a data segment being encoded in a time domain or in a freguency domain, a data segment
being encoded in the frequency domain having successive blocks of data representing successive
and overlapping blocks of time-domain data samples This data stream could, for example,
correspond to an audio stream, wherein some of the data blocks are encoded in the time domain
and other ones are encoded in the frequency domain Data blocks or segments which have been
encoded in the frequency domain, may represent time-domain data samples of overlapping data
blocks
The apparatus 100 comprises a time-domain decoder 110 for decoding a data segment-being
encoded in the time domain Furthermore, the apparatus 100 comprises a processor 120 for
processing the data segment being encoded in the frequency domain and output data of the time-
domain decoder 110 to obtain overlapping time-domain data blocks.
Moreover, the apparatus 100 comprises an overlap/add-combiner 130 for combining the
overlapping time-domain data blocks to obtain the decoded data segments of the time-
domain data stream
Fig lb shows another embodiment of the apparatus 100 In embodiments the processor 120 may
comprise a frequency-domain decoder 122 for decoding data segments being encoded in the
frequency domain to obtain frequency-domain data segments Moreover, in embodiments the
processor 120 may

comprise a time-domain to frequency-domain converter 124
for converting the output data of the time-domain decoder
110 to obtain converted frequency-domain data segments.
Furthermore, in embodiments the processor 120 may comprise
a frequency-domain combiner 126 for combining the
frequency-domain segments and the converted frequency-
domain data segments to obtain a frequency-domain data
stream. The processor 120 may further comprise a frequency-
domain to time-domain converter 128 for converting the
frequency-domain data stream to overlapping time-domain
data blocks which can then be combined by the overlap/add-
combiner 130.
Embodiments may utilize an MDCT filterbank, as for example,
used in MPEG-4 AAC, without any modifications, especially
without giving up the property of critical sampling.
Embodiments may provide optimum coding efficiency.
Embodiments may achieve a smooth transition to a time-
domain codec compatible with the established MDCT windows
while introducing no additional switching artifacts and
only a minimal overhead.
Embodiments may keep the time-domain aliasing in the
filterbank and intentionally introduce a corresponding
time-domain aliasing into the signal portions coded by the
time-domain codec. Thus, resulting components of the time-
domain aliasing can cancel each other out in the same way
as they do for two consecutive frames of the MDCT spectra.
Fig. lc illustrates another embodiment of an apparatus 100.
According to Fig. lc the frequency-domain decoder 122 can
comprise a re-quantization stage 122a. Moreover, the time-
domain to frequency-domain converter 124 can comprise a
cosine modulated filterbank, an extended lapped transform,
a low delay filterbank or a polyphase filterbank. The
embodiment shown in Fig. lc illustrates that the time-

domain to frequency-domain converter 124 can comprise an
MDCT 124a.
Furthermore, Fig. lc depicts that the frequency-domain
combiner 126 may comprise an adder 126a. As shown in Fig.
lc, the frequency-domain to time-domain converter 128 can
comprise a cosine modulated filterbank, respectively an
inverse MDCT 123a. The data stream comprising time-domain
encoded and frequency-domain encoded data segment may be
generated by an encoder which will be further detailed
below. The switching between frequency-domain encoding and
time-domain encoding can be achieved by encoding some
portions of the input signal with a frequency-domain
encoder and some input signal portions with a time-domain
encoder. The emoodiment of the apparatus 100 depicted in
Fig. lc illustrates the principle structure of a
corresponding apparatus 100 for decoding. In other
embodiments the re-quantization 122a and the inverse
modified discrete cosine transform 128a can represent a
frequency-domain decoder.
As indicated in Fig. lc for signal portions where the time-
domain decoder 110 takes over, the time-domain output of
the time-domain decoder 110 can be transformed by the
forward MDCT 124a. The time-domain decoder may utilize a
prediction filter to decode the time-domain encoded data.
Some overlap in the input of the MDCT 124a and thus some
overhead may be introduced here. In the following
embodiments will be described which reduce or minimize this
overhead.
In principle, the embodiment shown in Fig. lc also
comprises an operation mode where both codecs can operate
in parallel. In embodiments the processor 120 can be
adapted for processing a data segment being encoded in
parallel in the time domain and in the frequency domain. In
this way the signal can partially be coded in the frequency
domain and partially in the time domain, similar to a

layered coding approach. The resulting signals are then
added up in the frequency domain, compare the frequency-
domain combiner 126a. Nevertheless, embodiments may carry
out a mode of operation which is to switch exclusively
between the two codecs and only have a preferably minimum
number of samples where both codecs are active in order to
obtain best possible efficiency.
In Fig. lc, the output of the time-domain decoder 110 is
transformed by the MDCT 124a, followed by the IMDCT 128a.
In another embodiment, these two steps may be
advantageously combined into a single step in order to
reduce complexity. Fig. Id illustrates an embodiment of an
apparatus 100 illustrating this approach. The apparatus 100
shown in Fig. Id illustrates that the processor 120 may
comprise a calculator 129 for calculating overlapping time-
domain data blocks based on the output data of the time-
domain decoder 110. The processor 120 or the calculator 129
can be adapted for reproducing a property respectively an
overlapping property of the frequency-domain to time-domain
converter 128 based on the output data of the time-domain
decoder 110, i.e. the processor 120 or calculator 129 may
reproduce an overlapping characteristic of time-domain data
blocks similar to an overlapping characteristic produced by
the frequency-domain to time-domain converter 128.
Moreover, the processor 120 or calculator 129 can be
adapted for reproducing time-domain aliasing similar to
time-domain aliasing introduced by the frequency-domain to
time-domain converter 128 based on the output data of the
time-domain decoder 110.
The frequency-domain to time-domain converter 128 can then
be adapted for converting the frequency-domain data
segments providec by the frequency-domain decoder 122 to
overlapping time-domain data blocks. 'The overlap/add-
combiner 130 can be adapted for combining data blocks
provided by the frequency-domain to time-domain converter

Amended page 12
128 and the calculator 129 to obtain the decoded data segments
of the time-domain data stream.
The calculator 129 may comprise a time-domain aliasing stage
129a as it is illustrated in the embodiment shown in Fig. le.
The time-domain aliasing stage 129a can be adapted for time-
aliasing output data of the time-domain decoder to obtain the
overlapping time-domain data blocks.
For the time-domain encoded data a combination of the MDCT and
the IMDCT can make the process in embodiments much simpler in
both structure and computational complexity as only the
process of time-domain aliasing (TDA) remains in embodiments.
This efficient process can be based on a number of
observations. The windowed MDCT of the input segments of 2N
samples can be decomposed into three steps.
First, the input signal is multiplied by an analysis window.
Second, the result is then folded down from 2N samples to N
samples. For the MDCT, this process implies that the first
quarter of the samples is combined, i.e. subtracted, in time-
reversed order with the second quarter of the samples, and
that the fourth quarter of the samples is combined, i.e.
added, with the third quarter of the samples in time-reversed
order. The result is the time-aliased, down-sampled signal in
the modified second and third quarter of the signal,
comprising N samples.
Third, the down-sampled signal is then transformed using an
orthogonal DCT-like transform mapping N input to N output
samples to form the final MDCT output.
The windowed IMDCT reconstruction of an input sequence of N
spectral samples can likewise be decomposed into three steps.

First, the input sequence of N spectral samples is
transformed using an orthogonal inverse DCT-like transform
mapping N input to N output samples.
Second, the results unfolded from N to 2N samples by
writing the inverse DCT transformed values into the second
and third quarter of a 2N samples output buffer, filling
the first quarter with the time-reversed and inverted
version of the second quarter, and the fourth quarter with
a time-reverse version of the third quarter, respectively.
Third, the resulting 2N samples are multiplied with the
synthesis window to form the windowed IMDCT output.
Thus, a concatenation of the windowed MDCT and the windowed
IMDCT may be efficiently carried out in embodiments by the
sequence of the first and second steps of the windowed MDCT
and the second and third steps of the windowed IMDCT. The
third step of the MDCT and the first step of the IMDCT can
be omitted entirely in embodiments because they are inverse
operations with respect to each other and thus cancel out.
The remaining steps can be carried out in the time domain
only, and thus embodiments using this approach can be
substantially low in computational complexity.
For one block of MDCT and consecutive IMDCT, the second and
third step of the MDCT and the second and third step of the
IMDCT can be written as a multiplication with the following
sparse 2Nx2N matrix.


In other words, the calculator 129 can be adapted for
segmenting the output of the time-domain decoder 110 in
calculator segments comprising 2N sequential samples,
applying weights to the 2N samples according to an analysis
windowing function, subtracting the first N/2 samples in
reversed order from the second N/2 samples, and the last
N/2 samples in reversed order to the third N/2 samples,
inverting the second and third N/2 samples, replacing the
first N/2 samples with the time-reversed and inverted
version of the second N/2 samples, replacing the fourth N/2
samples with the time reversed version of the third N/2
samples, and applying weights to the 2N samples according
to a synthesis windowing function.
In other embodiments the overlap/add-combiner 130 can be
adapted for applying weights according to a synthesis
windowing function to overlapping time-domain data blocks
provided by the frequency-domain to time-domain converter
128. Furthermore, the overlap/add-combiner 130 can be
adapted for applying weights according to a synthesis
windowing function being adapted to the size of an
overlapping region of consecutive overlapping time-domain
data blocks.
The calculator 129 may be adapted for applying weights to
the 2N samples according to an analysis windowing function
being adapted to the size of an overlapping region of

consecutive overlapping time-domain data blocks and the
calculator may be further adapted for applying weights to
the 2N samples according to a synthesis window function
.being adapted to the size of the overlapping region.
In embodiments the size of an overlapping region of two
consecutive time-domain data blocks which are encoded in
the frequency-domain can be larger than the size of ' an
overlapping of two consecutive time-domain data blocks of
which one being encoded in the frequency domain and one
being encoded in the time domain.
, In embodiments, the size of the data segments can be
adapted to the size of the overlapping regions. Embodiments
may have an efficient implementation of a combined
MDCT/IMDCT processing, i.e. a block TDA comprising the
operations of analysis windowing, folding and unfolding,
and synthesis windowing. Moreover, in embodiments some of
these steps may be partially or fully combined in an actual
implementation.
.Another embodiment of an apparatus 100 as shown in Fig. If
illustrates that an apparatus 100 may further comprise a
bypass 140 for the processor 120 and the overlay/add-
combiner 130 being adapted for bypassing the processor 120
and the overlay/add-combiner 130 when non-overlapping
consecutive time-domain data blocks occur in data segments,
which are encoded in the time domain. If multiple data
segments are encoded in the time domain, i.e. no conversion
to the frequency domain may oe necessary for decoding
.consecutive data segments, they may be transmitted without
any overlapping. For these cases the embodiments as shown
in Fig. If may bypass the processor 120 and the
overlap/add-combiner 130. In embodiments the overlapping of
blocks can be determined according to the AAC-
specifications.

Fig. 2a shows an embodiment of an apparatus 200 for
generating an encoded data stream based on a time-domain
data stream, the time-domain data stream having samples of
a signal. The time-domain data stream could, for example,
correspond to an audio signal, comprising speech sections
and music sections or both at the same time. The apparatus
200 comprises a segment processor 210 for providing data
segments from the data stream, two consecutive data
segments having a first or a second overlapping region, the
second overlapping region being smaller than the first
overlapping region. The apparatus 200 further comprises a
time-domain encoder 220 for encoding a data segment in the
time domain and a frequency-domain encoder 230 for applying
weights to samples of the time-domain data stream according
to a first or a second windowing function to obtain a
windowed data segment, the first and second windowing
functions being adapted to the first and second overlapping
regions and for encoding the windowed data segment in the
frequency domain.
Furthermore, the apparatus 200 comprises a time-domain data
analyzer 240 for determining a transmission indication
associated with a data segment and a controller 250 for
controlling the apparatus such that for data segments
having a first transition indication, output data of the
time-domain encoder 220 is included in the encoded data
stream and for data segments having a second transition
indication, output data of the frequency-domain encoder 230
is included in the encoded data stream.
In embodiments the time-domain data analyzer 240 may be
adapted for determining the transition indication from the
time-domain data stream or from data segments provided by
the segment processor 210. These embodiments are indicated
in Fig. 2b. In Fig. 2b it is illustrated that the time-
domain data analyzer 240 may be coupled to the input of tne
segment processor 210 in order to determine the transition
indication from the time-domain data stream. In another

embodiment the time-domain data analyzer 240 may be coupled
to the output of the segment processor 210 in order to
determine the transition indication from the data segments.
In embodiments the time-domain data analyzer 240 can be
coupled directly to the segment processor 210 in order to
determine the transition indication from data provided
directly by the segment processor. These embodiments are
indicated by the dotted lines in Fig. 2b.
In embodiments the time-domain data analyzer 240 can be
adapted for determining a transition measure, the
transition measure being based on a level of transience in
the time-domain data stream or the data segments wherein
the transition indicator may indicate whether the level of
transience exceeds a predetermined threshold.
Fig. 2c shows another embodiment of the apparatus 200. In
the embodiments shown in Fig. 2c the segment processor 210
can be adapted for providing data segments with the first
and the second overlapping regions, the time-domain encoder
220 can be adapted for encoding all data segments, the
frequency-domain encoder 230 may be adapted for encoding
all windowed data segments and the controller 250 can be
adapted for controlling the time-domain encoder 220 and the
frequency-domain encoder 220 and the frequency-domain
encoder 230 such that for data segments having a first
transition indication, output data of the time-domain
encoder 220 is included in the encoded data stream and for
data segments having a second transition indication, output
data of the frequency-domain encoder 230 is included in the
encoded data stream. In other embodiments both output data
of the time-domain encoder 220 and the frequency-domain
encoder 230 may be included in the encoded data stream. The
transition indicator may be indicating whether a data
segment is rather associated or correlated with a speech
signal or with a music signal. In embodiments the
frequency-domain encoder 230 may be used for more music-
like data segments and the time-domain encoder 220 may be

used for more speech-like data segments. In embodiments
parallel encoding may be utilized, e.g. for a speech-like
audio signal having background music.
In the embodiment depicted in Fig. 2c, multiple
possibilities are conceivable for the controller 250 to
control the multiple components within the apparatus 200.
The different possibilities are indicated by dotted lines
in Fig. 2c. For example, the controller 250 could be
coupled to the time-domain encoder 220 and the frequency-
domain encoder 230 in order to choose which encoder should
produce an encoded output based on the transition
indication. In another embodiment the controller 250 may
control a switch at the outputs of the time-domain encoder
220 and the frequency-domain encoder 230.
In such an embodiment both the time-domain encoder 220 and
the frequency-domain encoder 230 may encode all data
segments and the controller 250 may be adapted for choosing
via said switch which is coupled to the outputs of the
encoders, which encoded data segment should be included in
the encoded data stream, based on coding efficiency,
respectively the transition indication. In other
embodiments the controller 250 can be adapted for
controlling the segment processor 210 for providing the
data segments either to the time-domain encoder 220 or the
frequency-domain encoder 230. The controller 250 may also
control the segment processor 210 in order to set
overlapping regions for a data segment. In other
embodiments the controller 250 may be adapted for
controlling a switch between the segment processor 210 and
the time-domain encoder 220, respectively the frequency-
domain encoder 230. The controller 250 could then influence
the switch so to direct data segments to either one of the
encoders, respectively to both. The controller 250 can be
further adapted to set the windowing functions for the
frequency-domain encoder 230 along with the overlapping
regions and coding strategies.

Moreover, in embodiments the frequency-domain encoder 230
can be adapted for applying weights of window functions
according to AAC specifications. The frequency-domain
encoder 230 can be adapted for converting a windowed data
segment to the frequency domain to obtain a frequency-
domain data segment. Moreover, the frequency domain encoder
230 can be adapted for quantizing the frequency-domain data
segments and, furthermore, the frequency-domain encoder 230
may be adapted for evaluating the frequency-domain data
segments according to a perceptual model.
The frequency-domain encoder 230 can be adapted for
utilizing a cosine modulated filterbank, an extended lapped
transform, a low-delay filterbank or a polyphase filterbank
to obtain the frequency-domain data segments.
The frequency-domain encoder 230 may be adapted for
utilizing an MDCT to obtain the frequency data segments.
The time-domain encoder 220 can be adapted for using a
prediction model for encoding the data segments.
In embodiments where an MDCT in the frequency-domain
encoder 230 operates in a so-called long block mode, i.e.
the regular mode of operation that is used for coding non-
transient input signals, compare AAC-specifications, the
overhead introduced by the switching process may be high.
This can be true for the cases where only one frame, i.e. a
length/framing rate of N samples, should be coded using the
time-domain encoder 220 instead of the frequency-domain
encoder 230.
Then all the input values for the MDCT may have to be
encoded with the time-domain encoder 220, i.e. 2N samples
are available at the output of the time-domain decoder 110.
Thus, an overhead of N additional samples could be
introduced. Figs. 3a to 3d illustrate some conceivable
overlapping regions of segments, respectively applicable

windowing functions. 2N samples may have to be coded with
the time-domain encoder 220 in order to replace one block
of frequency-domain encoded data. Fig. 3a illustrates an
example, where frequency-domain encoded data blocks use a
solid line, and time-domain encoded data uses a dotted
line. Underneath the windowing functions data segments are
depicted which can be encoded in the frequency domain
(solid boxes) or in the time domain (dotted boxes). This
representation will be referred to in Figs. 3b to 3d as
well.
Fig. 3a illustrates the case where data is encoded in the
frequency domain, interrupted by one data segment which is
encoded in the time domain, and the data segment after it
is encoded in the frequency domain again. In order to
provide the time-domain data which is necessary to cancel
the time-domain aliasing evoked by the frequency-domain
encoder 230, when switching from the frequency domain to
the time domain, half of a segment size of overlapping is
required, the same holds from switching back from the time
domain to the frequency domain. Assuming that the time-
domain encoded data segment in Fig. 3a has a size of 2N,
then at its start and at the end it overlaps with the
frequency-domain encoded data by N/2 samples.
In case more than one subsequent frames can be encoded
using the time-domain encoder 220, the overhead for the
time-domain encoded section stays at N samples. As it is
illustrated in Fig. 3b where two consecutive frames are
encoded in the time domain and the overlapping regions at
the beginning and the end of the time-domain encoded
sections have the same overlap as it was explained with
respect to Fig. 3a. Fig. 3b shows the overlap structure in
case of two frames encoded with time-domain encoder 220. 3N
samples have to be coded with the time-domain encoder 220
in this case.

This overhead can be reduced in embodiments by utilizing
window switching, for example, according to the structure
which is used in AAC. Fig. 3c illustrates a typical
sequence of Long, Start, 8Short and Stop windows, as they
are used in AAC. From Fig. 3c it can be seen that the
window sizes, the data segment sizes and, consequently, the
size of the overlapping regions change with the different
windows. The sequence depicted in Fig. 3c is an example for
the sequence mentioned above.
Embodiments should not be limited to windows of the size of
AAC windows, however, embodiments take advantage of windows
with different overlapping regions and also of windows of
different durations. In embodiments transitions to and from
short windows may utilize a reduced overlap as, for
example, disclosed in Bernd Edler, "Codierung von
Audiosignalen mit uberlappender Transformation und
adaptiven Fensterfunktionen", Frequenz, Vol. 43, No. 9, p.
252-256, September 1989 and Generic Coding of Moving
Pictures and1 Associated Audio: Advanced Audio Coding,
International Standard 13818-7, ISO/IEC JTC1/SC29/WG11
Moving Pictures Expert Group, 1997 may be used in
embodiments to reduce the overhead for the transitions to
and from the time-domain encoded regions, as it is
illustrated in Fig. 3d. Fig. 3d illustrates four data
segments, of which the first two and the last one are
encoded in the frequency domain and the third one is
encoded in the time domain. When switching from the
frequency domain to the time domain different windows with
the reduced overlapping size are used, therewith reducing
the overhead.
In embodiments the transition may be based on Start and
Stop windows identical to the ones used in AAC. The
corresponding windows for the transitions to and from the
time-domain encoded regions are windows with only small
regions of overlap. As a consequence, the overhead, i.e.
the number of additional values to be transmitted due to

the switching process decreases substantially. Generally,
the overhead may be Novi/2 for each transition with the
window overlap cf Novi samples. Thus, a transition with the
regular fully-overlapped window like an AAC with Novl = 1024
incurs an overhead of 1024/2 = 512 samples for the left,
i.e. the fade-in window, and 1024/2 = 512 samples for the
right, i.e. the fade-out window, transition resulting in a
total overhead of 1024 (= N) samples. Choosing a reduced
overlap window like the AAC Short block windows with
Novl=128 only results in an overall overhead of 128 samples.
Embodiments may utilize a filterbank in the frequency-
domain encoder 230 as, for example, the widely used MDCT
filterbank, however, other embodiments may also be used
with frequency-domain codecs based on other cosine-
modulated filterbanks. This may comprise the derivates of
the MDCT, such as extended lapped transforms or low-delay
filterbanks as well as polyphase filterbanks, such as, for
example, the one used in MPEG-l-Layer-1/2/3 audio codecs.
In embodiments efficient implementation of a forward/back-
filterbank operation may take into account a specific type
of window and folding/unfolding used in the filterbank. For
every type of modulated filterbank the analysis stage may
be implemented efficiently by a preprocessing step and a
block transform, i.e. DCT-like or DFT, for the modulation.
In embodiments the corresponding synthesis stage can be
implemented using the corresponding inverse transform and a
post processing step. Embodiments may only use the pre- and
post processing steps for the time-domain encoded signal
portions.
Embodiments of the present invention provide the advantage
that a better code efficiency can be achieved, since
switching between a time-domain encoder 220 and the
frequency-domain encoder 230 can be done introducing very
low overhead. In signal sections of subsequent time-domain
encoding only, overlap may be 'omitted completely in

embodiments. Embodiments of the apparatus 100 enable the
according decoding of the encoded data stream.
Embodiments therewith provide the advantage that a lower
coding rate can be achieved for the same quality of, for
example, an audio signal, respectively a higher quality can
be achieved with the same coding rate, as the respective
encoders can be adapted to the transience in the audio
signal.
Depending on certain implementation requirements of the
inventive methods, the inventive methods can be implemented
in hardware or in software. The implementation can be
performed using a digital storage medium, in particular a
disc, DVD or CD having electronically stored control
signals stored thereon, which corporate with the
programmable computer system such that the inventive
methods are performed. Generally, the present invention is,
therefore, a computer program product having a program code
stored on a macnine-readable carrier, the program code
being operative for performing the inventive methods when
the computer program product runs on a computer. In other
words, the inventive methods are, therefore, a computer
program having a program code for performing at least one
of the inventive methods when the computer program runs on
a computer.

Reference List
100 apparatus for decoding
110 time-domain decoder
120 processor
122 frequency-domain decoder
122a re-quantization
124 time-domain to frequency-domain converter
124a modified discrete cosine transform
126 frequency-domain combiner
126a adder
128 frequency-domain to time-domain converter
128a inverse modified discrete cosine transform
129 calculator
129a time-domain aliasing stage
130 overlap/add-combiner
200 apparatus for encoding
210 segment processor
220 time-domain encoder
230 frequency-domain encoder
240 time-domain data analyzer
250 controller
400 modified discrete cosine transform input
410 windows
420 inverse modified discrete cosine transform output
first window
425 inverse modified discrete cosine transform output
second window
430 final output

We Claim:
1. An apparatus for decoding (100) data segments representing a time-
domain data stream, one or more data segments being encoded in the
time domain, one or more data segments being encoded in the frequency
domain having successive blocks of data representing successive and
overlapping blocks of time-domain data samples, the apparatus (100)
comprising:
a time-domain decoder (110) for decoding a data segment being encoded
in the time domain;
a processor (120) for processing the data segments-being encoded in the
frequency domain and output data of the time-domain decoder (110) to
obtain time-domain data blocks such that time-domain data blocks
obtained based on subsequent data segments being encoded in the
frequency domain overlap, and
such that consecutive time-domain data blocks of which one is encoded in
the frequency domain and of which one is encoded in the time domain,
overlap; and
an overlap/add-combiner (130) for combining the overlapping time-
domain data blocks to obtain the decoded data segments of the time-
domain data stream;

wherein the overlap/add-combiner (130) is adapted to apply weights
according to synthesis windowing functions to overlapping time-domain
data blocks,
wherein the synthesis windowing function is adapted to a size of an
overlapping region of consecutive overlapping time-domain data blocks,
wherein a window with a reduced overlapping size is applied to a time-
domain data block encoded in the frequency domain when switching from
the frequency-domain to the time domain;
wherein a size of an overlapping region of two consecutive time-domain
data blocks which are encoded in the frequency-domain is larger than a
size of an overlapping region of two consecutive time-domain data blocks
of which one is encoded in the frequency-domain and one is encoded in
the time domain.
2. The apparatus as claimed in claim 1, wherein the processor (120) comprises a frequency-domain decoder (122) for decoding data segments
being encoded in the frequency domain to obtain frequency-domain data
segments.

3. The apparatus as claimed in claim 1, wherein the processor (120) is
adapted for processing a data segment being encoded in the time domain
and in the frequency domain in parallel.
4. The apparatus as claimed in claim 2, wherein the processor (120)
comprises a time-domain to frequency-domain converter (124) for
converting the output data of the time-domain decoder (122) to obtain
converted frequency-domain data segments.
5. The apparatus as claimed in claim 4, wherein the processor (120)
comprises a frequency-domain combiner (126) for combining the
frequency-domain data segments and the converted frequency-domain
data segments to obtain a frequency-domain data stream.
6. The apparatus as claimed in claim 5, wherein the processor (120)
comprises a frequency-domain to time-domain converter (128) for
converting the frequency-domain data stream to overlapping time-domain
data blocks.
7. The apparatus as claimed in claim 2, wherein the frequency domain
decoder (122) comprises a re-quantization stage.
8. The apparatus as claimed in claim 4, wherein the time-domain to
frequency-domain converter (124) comprises a cosine modulated
filterbank, an extended lapped transform, a low-delay filterbank, a
polyphase filterbank or a modified discrete cosine transform (124a).

9. The apparatus as claimed in claim 5, wherein the frequency-domain
combiner (126) comprises an adder (126a).
10.The apparatus as claimed in claim 6, wherein the frequency-domain to
time-domain converter (128) comprises a cosine modulated filterbank or
an inverse modified discrete cosine transform (128a).
11.The apparatus as claimed in claim 1, wherein the time-domain decoder
(110) is adapted for using a prediction filter to decode a data segment
encoded in the time domain.
12.The apparatus as claimed in claim 1, wherein the processor (120)
comprises a calculator (129) for calculating overlapping time-domain data
blocks based on the output data of the time-domain decoder (110).
13.The apparatus as claimed in claim 12, wherein the calculator (129) is
adapted for reproducing an overlapping property of the frequency-domain
to time-domain converter (128) based on the output data of the time-
domain decoder (110).
14.The apparatus as claimed in claim 13, wherein the calculator (129) is
adapted for reproducing a time-domain aliasing characteristic (129a) of
the frequency-domain to time-domain converter (128) based on the
output data of the time-domain decoder (110).

15.The apparatus as claimed in claim 6, wherein the frequency-domain to
time-domain converter (128) is adapted for converting the frequency-
domain data segments provided by the frequency-domain decoder (122)
to overlapping time-domain data blocks.
16. The apparatus as claimed in claim 15, wherein the overlap/add-combiner
(130) is adapted for combining the overlapping time-domain data blocks
provided by the frequency-domain to time-domain converter (128) and the
calculator (129) to obtain decoded data segments of the time-domain data
stream.
17. The apparatus as claimed in claim 8, wherein the calculator (129)
comprises a time-domain aliasing stage (129a) for time-aliasing output
data of the time-domain decoder (110) to obtain the overlapping time-
domain data blocks.
18.The apparatus as claimed in claim 12, wherein the calculator (129) is
adapted for segmenting the output of the time-domain decoder in
calculator segments comprising 2N sequential samples,
applying weights to the 2N samples according to an analysis window
function,

subtracting the first N/2 samples in reversed order from the second N/2
samples,
adding the last N/2 samples in reversed order to third N/2 samples,
inverting the second and third N/2 samples replacing the first N/2 samples
with the time-reversed and inverted version of the second N/2 samples,
replacing the fourth N/2 samples with the time-reversed version of the
third N/2 samples, and applying weights to the 2N samples according to a
synthesis windowing function.
19. The apparatus as claimed in claim 6, wherein the overlap/add-combiner
(130) is adapted for applying weights according to a synthesis windowing
function to overlapping time-domain data blocks provided by the
frequency-domain to time-domain converter (128).
20. The apparatus as claimed in claim 19, wherein the overlap/add-combiner
(130) is adapted for applying weights according to a synthesis windowing
function being adapted to a size of an overlapping region of consecutive
overlapping time-domain data blocks.

21.The apparatus as claimed in claim 20, wherein the calculator (129) is
adapted for applying weights to the 2N samples according to an analysis
windowing function being adapted to a size of an overlapping region of
consecutive overlapping time-domain data blocks and wherein the
calculator (129) is adapted for applying weights to the 2N samples
according to a synthesis windowing function being adapted to the size of
the overlapping region.
22.The apparatus as claimed in claim 1, wherein a size of an overlapping
region of two consecutive time-domain data blocks which are encoded in
the frequency domain is larger than a size of an overlapping region of two
consecutive time-domain data blocks of which one being encoded in the
frequency domain and one being encoded in the time domain.
23. The apparatus as claimed in claim 1, wherein the overlapping of data
blocks is being determined according to the AAC-specifications.
24. The apparatus as claimed in claim 1, comprising a bypass for the
processor (120) and the overlap/add-combiner (130), the bypass being
adapted for bypassing the processor and the overlap/add-combiner when
non-overlapping consecutive time-domain data blocks occur in data segments
which are encoded in the time domain.

25. Method for decoding data segments representing a time-domain data
stream, one or more data segments being encoded in the time domain,
one or more data segments being encoded in the frequency domain
having successive blocks of data representing successive and overlapping
blocks of time-domain data samples, comprising the steps of:
decoding a data segment being encoded in the time domain;
processing the data segment being encoded in the frequency domain and
output data of the time-domain decoder to obtain overlapping time-
domain data blocks such that time-domain data blocks obtained based on
subsequent data segments being encoded in the frequency domain
overlap, and
such that consecutive time-domain data blocks of which one is encoded in
the frequency domain and of which one is encoded in the time domain,
overlap; and
combining the overlapping time-domain data blocks to obtain the decoded
data segments of the time-domain data stream;
wherein weights according to synthesis window functions are applied to
the overlapping time-domain data blocks;
wherein the synthesis windowing function is adapted to a size of an
overlapping region of consecutive overlapping time-domain data blocks,

wherein a window with a reduced overlapping size is applied to a time-
domain data block encoded in the frequency domain when switching from
the frequency-domain to the time domain;
wherein a size of an overlapping region of two consecutive time-domain
data blocks which are encoded in the frequency-domain is larger than a
size of an overlapping region of two consecutive time-domain data blocks of which one is encoded in the frequency-domain and one is encoded in
the time domain.
26. An apparatus (200) for generating an encoded data stream based on a
time-domain data stream, the time-domain data stream having samples of
a signal, the apparatus (200) comprising
a segment processor (210) for providing data segments from the data
stream;
a time-domain encoder (220) for encoding a windowed data segment in
the time domain;
a frequency-domain encoder (230) for applying weights to samples of the
time-domain data stream according to a first or second windowing
function to obtain a windowed data segment, the first and second
windowing functions being adapted to overlapping regions of different

lengths, the frequency-domain encoder being adapted for encoding a
windowed data segment in the frequency domain;
a time-domain data analyzer (240) for determining a transition indication
associated with a data segment; and
a controller (250) for controlling the apparatus (200) such that for data
segments having a first transition indication output data of the time-
domain encoder (220) is included in the encoded data stream and for
data segments having a second transition indication, output data of the
frequency-domain encoder (230) is included in the encoded data stream;
wherein the controller (250) is adapted to set the windowing functions for
the frequency-domain encoder (230), such that a window (418) with'a
reduced overlapping size is used when switching from the frequency-
domain to the time-domain.
27.The apparatus as claimed in claim 26, wherein the controller is adapted to
set the windowing functions for the frequency-domain encoder such that a
size of an overlapping region of two consecutive windowed data segments
which are encoded in the frequency-domain, is larger than a size of an
overlapping region of two consecutive windowed data segments, of which
one is encoded in the frequency-domain and one is encoded in the time-
domain.

28. The apparatus as claimed in claim 26 or 27, wherein the time-domain data
"analyzer is adapted for determining the transition indication from the
time-domain data stream, the data segments or from data directly
provided by the segment processor.
29.The apparatus as claimed in claim 26 or 27, wherein the time-domain
data analyzer is adapted for determining a transition measure, the
transition measure being based on the level of transience in the time-
domain data stream or the data segment and wherein the transition
indicator indicates whether a level of transience exceeds a predetermined
threshold.
30. The apparatus as claimed in claim 26 or 27, wherein the segment
processor is adapted for providing data segments with overlapping regions
of different lengths, the time-domain encoder is adapted for encoding the
data segments,
the frequency-domain encoder is adapted for encoding the windowed data
segments, and
the controller is adapted for controlling the time-domain encoder and the
frequency-domain encoder such that for data segments having a first
transition indication output data of the time-domain encoder is included in
the encoded data stream and for windowed data segments having a

second transition indication output data of the frequency-domain encoder
is included in the encoded data stream.
31. The apparatus as claimed in claim 26 or 27, wherein the controller is
adapted for controlling the segment processor for providing the data
segments either to the time-domain encoder or the frequency-domain
encoder.
32.The apparatus as claimed in claim 26 or 27, wherein the frequency-
domain encoder is adapted for applying weights of windowing functions
. according to the AAC-specifications.
33.The apparatus as claimed in claim 26 or 27, wherein the frequency-
domain encoder is adapted for converting a windowed data segment to
the frequency domain to obtain a frequency-domain data segment.
34.The apparatus as claimed in claim 33, wherein the frequency-domain
encoder is adapted for quantizing the frequency-domain data segment.
35.The apparatus as claimed in claim 34, wherein the frequency-domain
encoder is adapted for evaluating the frequency-domain data segment
according to a perceptual model.

36. The apparatus as claimed in claim 35, wherein the frequency-domain
encoder is adapted for utilizing a cosine-modulated filterbank, an
extended lapped transform, a low-delay filterbank or a polyphase
filterbank to obtain the frequency-domain data segments.
37. The apparatus as claimed in claim 33, wherein the frequency-domain
encoder is adapted for utilizing a modified discrete cosine transform to
obtain the frequency-domain data segments.
38.The apparatus as claimed in claim 26 or 27, wherein the time-domain
encoder is adapted for using a prediction filter for encoding the data
' segments.
39. Method for generating an encoded data stream based on a time-domain
data stream, the time-domain data stream having samples of a signal,
comprising the steps of providing data segments from the data stream;
determining a transition indication associated with the data segments;
encoding a data segment in the time domain; and
applying weights to samples of the time-domain data stream according to
a first or a second windowing function to obtain a windowed data
segment, the first and second windowing functions being adapted to
overlapping regions of different lengths and encoding the windowed data
segment in the frequency domain and;

controlling such that for data segments having a first transition indication
output data being encoded in the time-domain is included in the encoded
data stream and for data segments having a second transition indication
output data being encoded in the frequency domain is included in the
encoded data stream;
wherein the windowing functions for the frequency-domain encoding are
set such that a window with a reduced overlapping size is used when
switching from the frequency-domain to the time-domain or from the
time-domain to the frequency domain.

ABSTRACT

TITLE: ENCODER, DECODER AND METHODS FOR ENCODING AND
DECODING DATA SEGMENTS REPRESENTING A TIME-DOMAIN DATA
STREAM
An apparatus (100) for decoding data segments representing a time-
domain data stream, a data segment being encoded in the time domain or
in the frequency domain, a data segment being encoded in the frequency
domain having successive blocks of data representing successive and
overlapping blocks of time-domain data samples The apparatus
comprises a time-domain decoder (110) for decoding a data segment
being encoded in the time domain and a processor (120) for processing
the data segment being encoded in the frequency domain and output data
of the time-domain decoder (110) to obtain overlapping time-domain data
blocks The apparatus further comprises an overlap/add-combiner (130)
for combining the overlapping time-domain data blocks to obtain a
decoded data segment of the time-domain data stream