Determining a budget for LPD/FD transition frame encoding
The present invention relates to the field of coding/decoding digital signals.
The invention advantageously applies to coding/decoding sounds which may
contain speech and music, either mixed together 5 or alternating.
In order to efficiently code the speech sounds at low rate, CELP type techniques
(“Code Excited Linear Prediction”) are recommended. In order to efficiently code the
music sounds, transform coding techniques are recommended instead.
CELP type coders are predictive coders. Their objective is to model speech
10 production from various elements: a short-term linear prediction to model the vocal tract, a
long-term prediction to model the vocal cord vibration during a voiced period, and an
excitation derived from a fixed dictionary (white noise, algebraic expectation) to represent
the “innovation” which could not be modeled.
The transform coders such as MPEG AAC, AAC-LD, AAC-ELD, or ITU-T
15 G.722.1 Annex C, for example, use critical sampling transforms in order to pack the signal
in the transform domain. “Critical sampling transform” refers to a transform for which the
number of coefficients in the transformed domain is equal to the number of time samples
in each analyzed frame.
A solution for efficiently coding a signal with mixed speech/music content consists
20 in selecting over time the best technique between at least two coding modes, one of CELP
type, the other of transform type.
It is for example the case for 3GPP AMR-WB+ and MPEG USAC codecs (for
“Unified Speech Audio Coding”). The applications aimed by AMR-WB+ and USAC are
not conversational, but corresponds to storing and disseminating services, without strong
25 constraints on the algorithmic delay.
The initial version of the USAC codec, called RM0 (Reference Model 0), is
described in the paper by M. Neuendorf et al., A Novel Scheme for Low Bitrate Unified
Speech and Audio Coding - MPEG RM0, 7-10 May 2009, 126th AES Convention. This
RM0 codec alternates between several coding modes:
30
• For speech type signals: LPD modes (for “Linear Predictive Domain”) comprising
two different modes derived from the AMR-WB+ coding:
An
ACELP mode
2
A
TCX (Transform Coded eXcitation) mode called wLPT (for “weighted Linear
Predictive Transform”) using an MDCT type transform (unlike the AMR-WB+ codec
which uses a FFT (“Fast Fourier transform”).
• For music type signals: FD mode (for “Frequency Domain”) using an MDCT
transform coding (for “Modified Discrete Cosine Transform”) 5 of MPEG AAC type
(for “Advanced Audio Coding”) over 1024 samples.
In the USAC codec, the transitions between LPD and FD modes are crucial for
ensuring a sufficient quality without switching flaws, knowing that each mode (ACELP,
TCX, FD) has a specific “signature” (in terms of artefacts) and that the FD and LPD modes
10 are different in nature – the FD mode is based on transform coding in the signal domain,
whereas the LPD modes use predictive linear coding in the perceptual weighted domain
with filter memories to correctly manage. Intermodal switching management in the USAC
RM0 codec is detailed in the paper by J. Lecomte et al., "Efficient cross-fade windows for
transitions between LPC-based and non-LPC based audio coding", 7-10 May 2009, 126th
15 AES Convention. As explained in this paper, the main difficulty is the transitions from
LPD to FD modes and vice versa. Only the case of CELP to FD transitions is considered
here.
In order to fully understand how it works, the principle of MDCT transform coding is
recalled through a typical example of development.
20 At the coder the MDCT transformation is typically divided into three steps, the signal
being split into frames of M samples prior to MDCT coding:
• Weighting the signal by a window here called “MDCT window” with length 2M.
• Time-domain aliasing for forming a block with length M.
• DCT transforming (for “Discrete Cosine Transform”) with length M
25 The MDCT window is divided into four adjacent portions with equal lengths M/2, here
called “quarters”.
The signal is multiplied by the analysis window, and then aliasing is performed: the
first quarter (windowed) is aliased (i.e. time reversed and overlapped) on the second
quarter and the fourth quarter is aliased on the third.
30 More precisely, time-domain aliasing a quarter on another is performed in the
following way: the first sample of the first quarter is added to (or subtracted from) the last
sample of the second quarter, the second sample of the first quarter is added to (or
subtracted from) the penultimate sample of the second quarter, and so on, until the last
3
sample of the first quarter which is added to (or subtracted from) the first sample of the
second quarter.
Therefore, from four quarters, 2 aliased quarters are obtained where each sample is the
result of a linear combination of 2 signal samples to code. This linear combination induces
time-5 domain aliasing.
The 2 aliased quarters are then jointly coded after the DCT transformation (of type IV).
For the following frame there is a shift by half of a window (i.e. 50% of overlap), the third
and fourth quarters of the preceding frame then become the first and the second quarter of
the current frame. After aliasing, a second linear combination of the same sample pairs is
10 sent like in the preceding frame, but with different weights.
At the decoder, after the DCT transformation, the decoded version of these aliased
signals is therefore obtained. Two consecutive frames contain the result for two different
aliasing events of the same quarters, i.e. for each sample pair there is the result for two
linear combinations with different but known weights: an equation system may therefore
15 be solved to obtain the decoded version of the input signal, time-domain aliasing may thus
be eliminated using two consecutive decoded frames.
Solving the mentioned equation systems is generally implicitly performed by opening,
multiplication by a synthesis window which is judiciously chosen, and then adding and
overlapping the common parts (without discontinuity due to quantization errors) between 2
20 consecutive decoded frames, indeed this operation behaves like a an overlap-add. When
the window for the first quarter or the fourth quarter is at zero for each sample, it is
referred to an MDCT transformation without time-domain aliasing in this part of the
window. In this case the smooth transition is not ensured by the MDCT transformation, it
must be made by other means such as for example an external overlap-add.
25 It should be noted that there are variants of implementations for the MDCT
transformation, in particular for defining the DCT transform, on the way of time-domain
aliasing the block to transform (for example, signs applied to the quarters aliased left and
right may be reversed, or the second and the third quarter may be aliased on the first and
fourth quarters respectively), etc. These variants do not change the MDCT analysis30
synthesis principle with sample block reduction by windowing, time-domain aliasing, then
transforming, and finally windowing, aliasing, and adding-overlapping.
In the case of the USAC RM0 coder described in the paper by Lecomte et al., the
transition between a frame coded by ACELP coding and a frame coded by FD coding, is
performed in the following way:
4
A transition window for the FD mode is used with an overlap to the left of 128
samples.
Time-domain aliasing on this overlap area is cancelled by introducing an artificial
time-domain aliasing to the right of the reconstructed ACELP frame. The MDCT windows
used for the transition has a size of 2304 samples and the DCT transformation 5 operates on
1152 samples whereas normally the FD mode frames are coded with a window with a size
of 2048 samples and a DCT transformation of 1024 samples. Thus the MDCT
transformation of the normal FD mode is not directly usable for the transition window, the
coder must also integrate a modified version of this transformation which complicates the
10 transition implementation for the FD mode.
This coding technique from the state-of-the-art has an algorithmic delay in the order of
100 to 200 ms. This delay is incompatible with conversational applications for which the
coding delay is generally in the order of 20 to 25 ms for speech coders for the mobile
applications (e.g. GSM EFR, 3GPP AMR and AMR-WB) and in the order of 40 ms for
15 conversational transform coders for videoconferencing (for example UIT-T G.722.1 Annex
C and G.719). Moreover, occasionally increasing the DCT transformation size (2304 vs
2048) causes a peak in complexity at the moment of transition.
To overcome these disadvantages, the international patent application
WO2012/085451, wherein the content is incorporated by reference to the present
20 application, proposes a new method of coding a transition frame. The transition frame is
defined as the transform coded current frame following a preceding frame coded by
predictive coding. According to the above-mentioned new method, part of the transition
frame, for example a sub-frame of 5 ms, in the case of CELP coding at 12.8 GHz, and two
extra CELP frames of 4 ms each, in the case of a CELP coding at 16 kHz, are coded by
25 predictive coding restricted with respect to predictive coding the preceding frame.
The restricted predictive coding consists in using the stable parameters of the preceding
frame coded by predictive coding, such as for example the linear prediction filter
coefficients and in only coding some minimum parameters for the extra sub-frame in the
transition frame.
30 As the preceding frame has not been coded with transform coding, cancelling timedomain
aliasing in the first part of the frame is impossible. The above-mentioned patent
application WO2012/085451 further proposes to modify the first half of the MDCT
window so as to not have time-domain aliasing in the first quarter which is usually aliased.
It is also proposed to integrate part of the overlap-add between the decoded CELP frame
5
and the decoded MDCT frame by modifying the coefficients of the analysis/synthesis
window. In reference to Figure 4e from the above-mentioned application, the chain-dotted
lines (lines alternating between dots and dashes) corresponds to MDCT coding aliasing
lines (top figure) and to the MDCT decoding aliasing lines (bottom figure). On the top
figure, the bold lines separate the frames of new samples at the coder input. 5 Coding a new
MDCT frame may be initiated when a frame as defined for new input samples is
completely available. It is important to note is that these bold lines at the coder do not
correspond to the current frame the two successive blocks of new samples arriving for each
frame: the current frame is in fact delayed by 8.75 ms corresponding to anticipation, named
10 “lookahead”. On the bottom figure, the bold lines separate the decoded frames at the
decoder output.
At the coder, the transition window is null until the aliasing point. Thus the coefficients
of the left part of the aliased window are identical to those of the non-aliased window. The
part between the aliasing point and the end of this transition (TR) CELP sub-frame
15 corresponds to a sinusoidal half-window. At the decoder, after opening, the same window
is applied to the signal. On the segment between the aliasing point and the start of the
MDCT frame, the window coefficients correspond to a sin² window. To ensure the
addition-overlap between the decoded CELP sub-frame and the signal originating from the
MDCT, only applying a cos² type window to the part of the CELP sub-frame in overlap
20 and to sum the latter with the MDCT frame is required. The method is of perfect
reconstruction.
However, the application WO2012/085451 provides allocating a bit budget Btrans for
coding the CELP sub-frame corresponding to the required budget for CELP coding a
classic frame, brought down to a single sub-frame. The remaining budget for transform
25 coding the transition frame is then insufficient and might lead to a quality decrease at low
rate.
The present invention improves the situation
For this purpose, a first aspect of the invention relates to a method of determining a
distribution of bits for coding a transition frame. This method is implemented in a
30 coder/decoder for coding/decoding a digital signal. The transition frame is preceded by a
predictive coded preceding frame and coding this transition frame comprises transform
coding and predictive coding a single sub-frame of the transition frame. The method
further comprises the following steps:
6
assigning
a bit rate for predictive coding the transition sub-frame, the bit rate being
equal to the minimum between the bit rate for transform coding the transition frame
and a first predetermined bit rate value;
determining
a first number of bits allocated for predictive coding the transition subframe
5 for the bit rate; and
calculating
a second number of bits allocated for transform coding the transition
frame from the first number of bits and a number of bits available for coding the
transition frame.
The predictive coding bit rate is thus curbed by a maximum value. The number of
10 bits allocated for predictive coding depends on this bit rate. Since the weaker the bit rate,
the weaker the number of bits allocated for coding is, a minimum remaining budget for
transform coding the transition frame is guaranteed.
Moreover, the number of bits allocated for predictive coding the sub-frame is
optimized with respect to the transform coding bit rate. Indeed, if the bit rate for transform
15 coding the transition frame is lower than the first predetermined value, the bit rate for
predictive coding and the bit rate for transfer coding are identical. Signal coherence thus
generated is therefore improved which simplifies the subsequent steps of coding (channel
coding) and processing the received frames at the decoder.
In another embodiment, the coder/decoder comprises a first core working, for
20 predictive coding/decoding a signal frame, at a first frequency, and a second core working,
for predictive coding/decoding a signal frame, at a second frequency. The first
predetermined bit rate value depends on the core selected from the first and second cores
for coding/decoding the predictive coded preceding frame.
The working frequency of the coder/decoder core has an influence on the number
25 of bits required for correctly representing the input digital signal. For example, for some
working frequencies, additional bits must be provided for coding frequency bands which
are non-directly processed by the core.
In an embodiment, when the first core has been selected for coding/decoding the
predictive coded preceding core, the assigned bit rate also equal to the maximum between
30 the bit rate for the transform coded transition frame and the second predetermined bit rate
value, the second value being lower than the first value. Thus, a minimum bit rate is
guaranteed in order to prevent rates differences being too large between the different coded
frames.
7
In another embodiment, the digital signal is decomposed into at least one frequency
low band and one frequency high band. In this situation, the first calculated number of bits
is assigned for predictive coding the transition frame for the frequency low band. A third
predetermined number of bits is thus allocated for coding the transition sub-frame for the
frequency high band. Moreover, the second number of bits allocated for transfer 5 coding the
transition frame is then further determined from the third predetermined number of bits.
Thus, it is possible to efficiently code the whole frequency spectrum of the input signal
without sacrificing the quality of signal restored upon decoding.
In an embodiment, the number of bits available for coding the transition frame is
10 fixed. This reduces the complexity of the coding steps.
In another embodiment, the second number of bits is equal to the fixed number of
bits for coding the transition frame minus the first number of bits minus the third number
of bits. The final determination of the distribution of bits in the transition frame is thus
limited to subtracting the entire values which simplifies coding.
15 Alternatively, the second number of bits is equal to the fixed number of bits for
coding the transition frame minus the first number of bits minus the third number of bits
minus a first bit minus a second bit. The first bit indicates whether a low-pass filtering is
performed during the determination of predictive coding parameters for the transition subframe,
the parameters being relative to the tonal lead time. The second bit indicates the
20 frequency used by the coder/decoder core for predictive coding/decoding the transition
sub-frame. Such indication allows more flexible coding.
A second aspect of the invention relates to a method of coding a digital signal in a
coder able to code signal frames according to predictive coding or according to transform
coding, comprising the following steps:
25  Coding a preceding frame of digital signal samples according to predictive coding;
 Coding a current frame of digital signal samples in a transition frame, coding the
transition frame comprising transform coding and predictive coding a single subframe
of the transition frame, coding the current frame comprising the following
sub-steps:
30 - determining the distribution of bits via the method according to the first
aspect of the invention;
- transform coding the transition frame on the second number of allocated
bits;
8
- predictive coding the transition sub-frame on the first number of allocated
bits.
Determining the distribution of bits comprised in the transition frame is thus
determined prior to coding. As described below, determining the distribution of bits is
reproducible at the decoder which prevents an explicit transmission 5 of information about
this distribution.
Moreover, this coding guarantees a balanced distribution between predictive coding
and transform coding within this transition frame.
In an embodiment, predictive coding comprises generating determined predictive
10 coding parameters for the bit rate assigned during the distribution of bits in the transition
frame. The use of such predictive parameters allows optimising the ratio between the bit
rate assigned for predictive coding and the rate remaining assigned for transform coding,
and therefore optimizing the quality of the reconstructed signal. Indeed, at constant quality,
the number of bits attributed for this predictive parameter or another may vary in non15
linear proportions with respect to the bit rate assigned for predictive coding.
In another embodiment, predictive coding comprises generating predictive coding
parameters restricted with respect to predictive coding the preceding frame by reusing at
least one predictive coding parameter of the preceding frame. Thus, upon decoding,
additional information is extracted from the preceding frame to complete decoding the
20 transition sub-frame to decode. This reduces the number of bits that must be reserved for
predictive coding the transition sub-frame.
The combination of reusing parameters from a preceding frame and assigning the
bit rate for transform coding the transition frame allows ensuring a coherent transition at
low-cost.
25 A third aspect of the invention relates to a method of decoding a digital signal
coded by predictive coding and transform coding, comprising the steps of:
 predictive decoding a preceding frame of digital signal samples coded according to
predictive coding;
 decoding a transition frame coding a current frame of digital signal samples,
30 including the transition frame comprising transform coding and predictive coding a
single sub-frame of the transition frame, comprising the sub steps of:
determining
the distribution of bits by the method according to the first
aspect of the invention;
9
predictive
decoding the transition sub-frame on the first number of allocated
bits;
transform
decoding the transition frame on the second number of allocated
bits.
As mentioned above, the method of determining the distribution 5 of bits in the
transition frame is directly reproducible at the decoder. Indeed, the distribution of bits is
determined only from the bit rate from the transform coded part of the transition.
Therefore, no extra bit is necessary for implementing the step of determining the
distribution of bits and bandwidth savings are therefore made.
10 A fourth aspect of the invention further aims for a computer program comprising
instructions for implementing the method according to the aspects of the invention
described above, when these instructions are executed by processor.
A fifth aspect of the invention relates to a device for determining a distribution of
bits for coding a transition frame, this device being implemented in a coder/decoder for
15 coding/decoding a digital signal, the transition frame being preceded by a predictive coded
preceding frame, coding the transition frame comprising transform coding and predictive
coding a single sub-frame of the transition frame, the number of bits for coding the
transition frame being fixed, the device comprising a processor arranged for performing
the following operations:
20 assigning
a bit rate for predictive coding the transition sub-frame, said bit
rate being equal to the minimum between the bit rate for transform coding
the transition frame and a first predetermined bit rate value;
determining
a first number of allocated of bits allocated for predictive
coding the transition sub-frame for the bit rate;
25 calculating
a second number of bits allocated for transform coding the
transition frame from the first required number of bits for coding the coding
parameters and the fixed number of bits for coding the transition frame.
A sixth aspect of the invention further aims for a coder able to code frames for a
digital signal according to predictive coding or according to transform coding, comprising :
30  a device according to the fifth aspect of the invention;
 a predictive coder comprising a processor arranged for performing the following
operations:
coding
a preceding frame of digital signal samples according to predictive coding;
10
predictive
coding a single sub-frame comprised in a transition frame coding a
current frame of digital signal samples, coding the transition frame comprising
transform coding and predictive coding the sub-frame, the processor being arranged
for performing the predictive coding operation for the transition sub-frame on the first
number 5 of allocated bits;
 a transform coder comprising a processor arranged for transform coding the
transition frame on the second number of allocated bits.
A seventh suspect of the invention further aims for a decoder for a digital signal coded
by predictive coding and transform coding, comprising:
10  a device according to the fifth aspect of the invention;
 a predictive decoder comprising a processor arranged for performing the following
operations:
predictive
decoding a preceding frame of digital signal samples coded according to
predictive coding;
15 predictive
decoding a single sub-frame comprised in a transition frame coding a
current frame of digital signal samples, coding the transition frame comprising
transform coding predictive coding the sub-frame, the processor being arranged for
performing the operation of predictive decoding the transition sub-frame and the
first number of allocated bits;
20  a transform decoder comprising a processor arranged for transform coding the
transition frame on the second number of allocated bits.
Other features and advantages of the invention will appear upon examining the detailed
description below, and the accompanying drawings on which:
Figure
1 illustrates an audio coder according to an embodiment of the invention;
25 Figure
2 is a diagram illustrating the steps of the coding method, implemented by
the audio coder of Figure 1, according to an embodiment of the invention;
Figure
3 shows a transition between the CELP and MDCT frames according to an
embodiment of the invention;
Figure
4 is a diagram illustrating the steps of a method of determining a distribution
30 of bits for coding a transition frame, according to an embodiment of the invention;
Figure
5 illustrates an audio decoder according to an embodiment of the invention;
Figure
6 is a diagram illustrating the steps of a method of decoding, implemented
by the audio decoder of Figure 5, according to an embodiment of the invention;
11
Figure
7 illustrates the device for determining the distribution of bits in a transition
frame according to an embodiment of the invention.
Figure 1 illustrates an audio coder 100 according to an embodiment of the invention.
Figure 2 is a diagram illustrating the steps of a method of coding, implemented by the
audio coder 100 of Figure 1, according to an embodiment 5 of the invention.
The coder 100 comprises a receiving unit 101 for receiving, at step 201, an input
signal samples at a given frequency fs (for example 8, 16, 32, or 48 kHz) and decomposed
into sub-frames, for example of 20 ms.
Upon receiving a current frame, a pre-processing unit 102 is able to select, at step 202,
10 the coding mode which is most adequate for coding the current frame, between at least one
LPD mode and one FD mode. In the following description, it is considered, for illustrative
purposes, that MDCT coding is used for the FD mode and that CELP coding is used for the
LPD mode. There is no restriction on the coding techniques employed for the LPD and FD
modes respectively. Thus, modes in addition to the CELP and MDCT modes may be used
15 for example, CELP coding may be replaced with another type of predictive coding, the
MDCT transform may be replaced with another type of transform.
It is assumed here that the frame type is explicitly transmitted via the block 206, with
for example fixed length coding indicating the mode chosen from a predefined list. In
variants of the invention, this coding for the mode chosen in each frame may be of variable
20 length. It is also provided that the CELP coding type (12.8 or 16 kHz) may be explicitly
transmitted through a bit so as to facilitate decoding the transition frame.
Step 203 verifies that CELP decoding has been selected at step 202. In cases where the
LPD mode is selected, the signal frame is transmitted to a CELP coder 103 for coding a
CELP frame at step 204. The CELP coder may use two “cores” working at two respective
25 internal sampling frequencies, for example fixed at 12.8 kHz and 16 kHz, which require
the use of sampling of the entry signal (at frequency fs) at an internal frequency of 12.8 or
16 kHz. Such re-sampling may be implemented in a re-sampling unit in the pre-processing
block 102 or in the CELP coder 103. The frame is then predictive coded by the CELP
coder 103 by deducting the CELP parameters generally depending on a signal
30 categorisation. The CELP parameters typically include LPC coefficients, a fixed and
adaptive gain vector, an adaptive dictionary vector, a fixed dictionary vector. This list may
also be modified based on a signal category in the frame, such as in UIT-T G.718 coding.
The parameters thus calculated may then be quantified, multiplexed, and transmitted at
12
step 206 to the decoder by a transmitting unit 108. The CELP coding parameters, such as
the LPC coefficients, the fixed and adaptive gain vector, the adaptive dictionary vector, the
fixed dictionary vector, and the CELP decoder states may further be memorised, at step
205, in a memory 107 in cases where the frame following the current frame is an MDCT
5 transition frame.
As explained below, a band extension may also be performed with coding associated
with the high band when the current frame is of CELP type.
In cases where MDCT coding has been selected by the unit 102 at step 203, it is
verified at step 207 that the frame preceding the current frame has been MDCT transform
10 coded. In cases where the frame preceding the current frame has been MDCT transform
coded, the current frame is transmitted to the MDCT coder 105 directly, for MDCT
transform coding the current frame at step 208. The MDCT coder may code a frame
covering 28.75 ms of non-re-sampled signal, including 20 ms of frame and 8.75 ms of
lookahead for example. There is no restriction on the MDCT window size. Furthermore, a
15 delay corresponding to the CELP coder delay due to the re-sampling of the input signal, is
applied to the frame coded by the MDCT coder, in such a way that the MDCT and CELP
frames are synchronized. Such delay at the coder may be of 0.9375 ms according to the resampling
type before CELP coding. The MDCT transform coded frame is transmitted to
the decoder at step 206.
20 In cases where MDCT coding is selected by the unit 102, and in cases where the frame
preceding the current frame has been predictive coded, the current frame a transition frame
and is transmitted to a transition unit 104. As described in the following, the MDCT
transition frame comprises an extra CELP sub-frame.
The transition unit 104 is able to implement the following steps:
25 anticipating,
at step 209, the budget of bits required for coding the transition CELP
sub-frame so as to define the budget available for MDCT coding the current frame. As
detailed in the following, the budget may depend on the current frame rate.
Furthermore, the budget may be evaluated depending on the CELP core used. In order
to preserve a sufficient bit budget for not degrading the quality of MDCT coding, the
30 invention may provide confining the coding rate for the CELP sub-frame. For this
purpose, it comprises a device for determining the distribution of bits in a transition
frame, such as the device 700 of Figure 7;
modifying,
at step 210, the MDCT window used at the coder in compliance with
Figure 3 described below;
13
zeroing
the MDCT transformation memory since the preceding frame is a CELP
frame, at step 207 – in the same manner the MDCT memory may be ignored in MDCT
decoding.
In an embodiment, at least one of these steps is performed by the transition frame
coding unit 106 5 described below.
The transition MDCT frame is coded by the MDCT coder 105, at step 212, as
described in the following, and based on a budget of bits allocated at step 209. The extra
CELP sub-frame is also coded by a CELP coder 103, at step 213, as described in the
following in reference to Figure 3, and depending on a bit budget allocated at step 209.
10 CELP coding may be performed before or after MDCT coding.
Figure 3 shows the transition between CELP and MDCT frames at the coder, before
coding, and at the decoder, before decoding.
A frame to code 301 is received at the coder 100 and is coded by the CELP coder 103.
A current frame 302 is then received by input of the coder 100 to be MDCT transform
15 coded. It is thus a transition frame. The following frame 303 received by input of the coder
is also MDCT transform coded. According to the invention, the following frame 303 may
be coded by CELP coding and there is no restriction on coding used for the following
frame 303.
An asymmetric MDCT window 304 may be used for coding the current frame. This
20 window 304 shows a rising edge 307 of 14.375 ms, a level with a gain at 1 of 11.25 ms, a
falling edge 309 of 8.75 ms corresponding to the lookahead, and a null part 310 of 5.265
ms. Adding the null part 310 allows reducing the lookahead and thus the corresponding
delay. In an embodiment, the form of this MDCT analysis window for MDCT coding is
modified for example for further reducing the lookahead or for using a symmetric window
25 with examples given in the patent application WO2012/085451.
The dashed lines 312 represent the medium of the MDCT window 304. On both sides
of the line 312, the 10 ms quarters of the MDCT window 212 are aliased as described in
the introductory part. The continuous line 311 indicates the aliasing area between the first
and second quarters of the MDCT window 304. The MDCT window of the following
30 frame 303 is referenced 306 and shows an overlap-add area with the MDCT window 304
corresponding to the falling edge 309 of the MDCT window 304.
An MDCT window 305 theoretically represents the window which will be applied to
the preceding window if it has been MDCT transform coded. However, the preceding
14
frame 301 being coded by the CELP coder 103, it is necessary, in order to allow opening
of the first part of the MDCT transform coded frame at the decoder, that the window is null
in the first quarter (since the second part of the preceding MDCT frame is not available).
For this purpose, the MDCT window 304 is modified by an MDCT window 313
having a first quarter at zero, allowing time-domain aliasing in the first 5 part of the MDCT
frame at the decoder.
At the decoder, the analysis windows 304, 305, 306, and 313 correspond to synthesis
windows 324, 325, 326, and 327 respectively. This synthesis window is therefore timereversed
with respect to the corresponding analysis window. In variants of the invention,
10 the analysis and synthesis windows may be identical, of sinusoidal type or other.
A first frame 320 of new samples coded by CELP coding is received at the decoder. It
corresponds to the coded version of this CELP frame 301. It is here recalled that the
decoded frame is shifted by 8.75 ms with respect to the frame 320.
The coded version of the transition frame 302, is then received (references 321 and
15 222 forming a complete frame). Between the end of the CELP frame 320 and the start of
the rising edge of the synthesis window 327 (corresponding to the aliasing line), a gap is
created. In the particular example represented here, a quarter of the MDCT window being
10 ms and the null part of the synthesis window MDCT 324 covering this CELP frame 220
being 5.625 ms (corresponding to the part 310 of the MDCT analysis window 204), the
20 gap is 4.275 ms. Furthermore, to ensure that a satisfying overlap-add length with the start
of the non-null part of the MDCT window 327, the delay between this CELP frame 320
and the start of the MDCT window 327 is prolonged to the required length. In the
following example, for illustrative purposes, a satisfying overlap-add length of 1.875 ms is
considered, the above-mentioned delay (corresponding to a missing signal length) thus
25 being brought up to 6.25 ms, as represented by reference 321 on Figure 2.
It should be noted that the signal frames represented on Figure 3 may contain signals
at different sampling frequencies of 12.8 or 16 kHz in cases of CELP coding/decoding and
of fs in cases of MDCT coding/decoding; however at the decoder, after re-sampling of the
CELP synthesis and time shift of the MDCT synthesis, the frames remain synchronised
30 and the representation of Figure 3 remains exact.
As previously mentioned the application WO2012/085451 proposes to code an extra
CELP sub-frame of 5 ms at the start of the MDCT transition frame, in cases of 12.8 kHz
CELP coding, and two extra CELP frames of 4 ms each at the start of the MDCT transition
frame, in cases of 16 kHz CELP coding.
15
In cases of 12.8 kHz, the 6.25 ms delay is not filled and the overlap-add is affected:
there are only 0.625 ms of overlap-add at the decoder, which is insufficient.
In the case of 16 kHz, two extra CELP sub-frames are coded at the start of the
transition frame, which only leaves a very little budget for coding the transition MDCT
frame and may lead to a significant quality decrease 5 at low rate.
In order to overcome these disadvantages, the present invention may provide coding a
single extra CELP sub-frame at 12.8 or 16 kHz by the CELP coder 103. The extra samples
are generated at the decoder, as detailed in the following, in order to generate the missing
signal on the above-mentioned 6.25 ms length.
10 In order to code the transition CELP sub-frame, the unit 106 may reuse at least one
CELP parameter of the preceding CELP frame. For example, the unit 106 may reuse the
linear prediction coefficient A(z) of the preceding CELP sub-frame as well as the energy
from the preceding frame innovation (stored in the memory 107 such as previously
described) in order to code only the adaptive dictionary vector, the adaptive gain, the fixed
15 gain, and the fixed dictionary vector of the transition CELP sub-frame. Thus, the extra
CELP sub-frame may be coded with the same core (12.8 kHz or 16 kHz) as the preceding
CELP frame.
A transition frame coding unit 106 ensures coding the transition frame according to
the invention. The invention may further provide the insertion by the unit 106 in the bit
20 flow of an extra bit indicating that the coded frame 322 is a transition frame, however in
general cases this transition frame indication may also be transmitted in the global
indication of the current frame coding mode, without taking extra bits.
The invention may further provide that this unit 116 codes the signal high band at
steps 204 and 214 (method of so-called “band extension”), when the latter is required, with
25 a fixed budget since the sampling frequency of the synthesis signal at the decoder is not
necessarily identical to the CELP core frequency.
For this purpose, the coding unit of the transition frame 106 may implement the
following steps:
filtering
the CELP preceding frame and the CELP sub-frame of the transition frame
30 by a high-pass filter in order to preserve the high part of the spectrum (above the
frequency corresponding to the used CELP core, i.e. above 6.4 or 8 kHz). Such
filtering may be implemented by a filter with finite impulse response, FIR, from the
CELP coder 103;
16
searching
correlation between the filtered part of the original transition CELP subframe
and the filtered preceding CELP frame in order to estimate a delay parameter
and then a gain (amplitude difference between the signal corresponding to the filtered
sub-frame and the signal predicted by applying the delay);
coding
the delay parameter and the above-mentioned gain using 5 for example scalar
quantification (for example, the delay may be coded over 6 bits and the gain over 6
bits).
The above-mentioned step 209 is illustrated in more details in reference to Figure 4
which is a diagram illustrating the steps of a method of determining a distribution of bits
10 for transition coding according to an embodiment of the invention. The above-mentioned
method is performed in the same manner at the coder and at the decoder, but is shown, for
illustrative purposes only, on the coder side.
At step 400, the total rate (in bit/s), noted core_brate, which may be allocated to
coding the current framed is fixed as being equal to the output rate of the MDCT coder.
15 The duration of the frame being considered in this example as 20 ms, the number of frames
per second is 50 and the total budgets in bits is equal to core_brate / 50. The total budget
may be fixed, in the case of a fixed rate coder, or variable, in the case of a variable rate
coder when adapting to the coding rate is implemented. In the following, an num_bits
variable is used, initialised at the value of core_brate / 50.
20 At step 401, the transition unit 104 determines the CELP core, from at least two CELP
cores, which has been used for coding this preceding CELP frame. In the following
example, two CELP cores are considered, working at frequencies of 12.8 kHz and 16 kHz
respectively. Alternatively, a single CELP core is implemented upon coding and/or upon
decoding.
25 In the case where the CELP core used for the preceding CELP frame has a 12.8 kHz
frequency, the method comprises a step 402 of assigning a bit rate, labelled cbrate, for
CELP coding the transition sub-frame, the bit rate being equal to the minimum between the
bit rate for MDCT coding of the transition frame and a first predetermined bit rate value.
The first predetermined value may be fixed at 24.4 kbit/s for example, which allows to
30 ensure a satisfying bit budget for transfer coding.
Thus, cbrate = min(core_bitrate, 24400). This limitation is equivalent to curbing the
operation of the restricted CELP coding limited to an extra sub-frame with the coded
CELP parameters as if they were coded by CELP coding with at most 24.40 kbit/s.
17
At an optional step 403, the assigned bit rate is compared to a 11.60 kbit/s CELP bit
rate. If the assigned bit rate is higher, a bit may be reserved for coding a bit indication for
low pass filtering the adaptive dictionary (such as for example for AMR-WB coding at
rates higher or equal to 12.65 kbit/s). The num_bits variable is updated:
num_bits 5 := num_bits – 1
At step 404, a first number of bits, labelled budg1, is allocated for predictive coding
the additional CELP sub-frame. The first number of bits budg1 represents the number of
bits representing the CELP parameters used for coding the CELP sub-frame. As previously
detailed, coding the CELP sub-frame may be restricted in that a restricted number of CELP
10 parameters is used, some parameters used for coding the preceding CELP frame being
reused advantageously.
For example, only the excitation may be modelled for coding the extra CELP subframe,
and bits are thus reserved only for the fixed dictionary vector, for the adaptive
dictionary vector, and for the gain vector. The number of bits attributed to each of these
15 parameters is deduced from the bit rate assigned for coding this extra CELP sub-frame at
step 402. For example, Table 1/G722.2 – Distribution of bits for the AMR-WB coding
algorithm for a 20 ms frame, originating from the July 2003 version of the G.722.2 of the
ITU-T, gives examples of bit allocations by a CELP parameter depending on the assigned
bit rate.
20 In the previous example, where coding the sub-frame is restricted, budg1 corresponds
to the sum of bits attributed to the adaptive dictionary, to the fixed dictionary, and to the
gain vector respectively. For example, for an assigned bit rate of 19.85 kbit/s, by referring
to the above-mentioned Table1/G722, 9 bits are allocated to the fixed dictionary (tonal lead
time), and 7 bits are allocated to the gain vector (directory gain). In this case, budg1 is
25 equal to 88 bits.
The num_bits variable may thus be updated:
num_bits := num_bits – budg1
The invention may also provide taking into account the frame categories in the
allocation of bits to the CELP parameters. For example, the G.718 norm of the ITU-T, in
30 its June 2008 version, sections 6.8 and 8.1, gives the budgets to allocate to each CELP
parameter depending on categories, or modes such as non-voiced mode (UC), the voiced
mode (VC), the transition mode (TC), and the generic mode (GC), and depending on the
allocated bit rate (layer1 or layer2, corresponding to rates of 8 kbit/s and 8+4 kbit/s,
respectively). The coder G.718 is a hierarchic coder, but it is possible to combine the
18
CELP coding principles using a G 718 categorisation with the multi-rate allocation of
AMR-WB.
If it has been determined at step 401 that the CELP core used for the preceding CELP
frame has a 16 kHz frequency, the method comprises a step 405 of assigning a bit rate,
labelled cbrate, for CELP coding the transition sub-frame, the bit rate 5 being equal to the
minimum between the bit rate for MGCT coding the transition frame and a first
predetermined value of the bit rate. In the case of the 16 kHz core, the first predetermined
value may be fixed at 22.6 kbit/s for example, which allows to ensure a satisfying bit
budget for transform coding. Thus, the first predetermined value depends on the CELP
10 core used for code the preceding CELP frame. Furthermore, for coding the 16 kHz core,
threshold values may be applied when assigning a bit rate to CELP coding. Thus, the
assigned bit rate is further equal to the maximum between the bit rate for the transform
coded transition frame and at least one predetermined second bit rate value, the second
value being lower than the first value. The second predetermined value of the trade may for
15 example be 14.8 kbit/s. Thus, if the bit rate for the transform coded transition frame is
lower than 14.8 kbit/s, the bit rate assigned for CELP coding the transition sub-frame may
be 14.8 kbit/s.
In a complementary embodiment, if the bit rate for the transform coded transition
frame is lower than 8 kbit/s, the assigned rate may be 8 kbit/s.
20 Thus, according to this complementary embodiment, the following algorithm is
obtained:
If core_bitrate ≤ 8000
cbrate = 8000
Otherwise if core_bitrate ≤ 14800
25 othercbrate = 14800
Otherwise
cbrate = min(core_bitrate , 22600)
End if
30 At an optional step 407, the assigned bit rate is compared to a CELP bit rate of
11.60 kbit/s. If the assigned bit rate is higher, a bit may be reserved for coding a low-pass
filtering bit indication of the adaptive dictionary. The num_bits variable is updated:
num_bits := num_bits – 1
At step 408, in the same manner as that of step 404, a first number of bits budg1 is
35 allocated for predictive coding the extra CELP sub-frame, and budg1 depends on the bit
rate assigned for CELP coding the transition sub-frame.
19
At step 410 which is common to coding at various core frequencies, a second
number is allocated for transform coding the transition frame, labelled budg2, is calculated
from the first number of bits budg1 the total number of bits of the transition frame.
Regarding the calculations above, budg2 is equal to the num_bits variable. Generally, the
mode of the transition current frame is here assumed to be imputed 5 to the MDCT coding
budget, this information is thus not explicitly taken into account.
The preceding steps may have been implemented for coding a frequency low band
of the transition sub-frame, in cases where the audio signal is decomposed into at least one
frequency low band and one frequency high band. At an optional step 409 preceding step
10 410, also common to coding at different core frequencies, the method may comprise
allocating a third predetermined number of bits, labelled budg3 for coding the frequency
high band of the transition sub-frame. In this case, the second number of bits budg2 is
calculated both from the first number of bits budg1 and the third number of bits budg3.
As previously explained, coding the frequency high band (or extending the band) of
15 the transition sub-frame may be based on a correlation between the preceding frame of the
audio signal and the transition sub-frame. For example, coding the frequency high band
may be decomposed into two steps.
In the first step, the preceding frame and the current frame of the audio signal are
filtered by a high pass filter to only keep the higher part of the spectrum. The high part of
20 the spectrum may correspond to frequencies higher than that of the CELP core used. For
example, if the CELP core used is the 12.8 kHz CELP core, the high band corresponds to
the audio signal for which the frequencies lower than 12.8 kHz have been filtered. Such
filtering may be implemented by means of an FIR filter.
In a second step, searching a correlation between the filtered parts of the preceding
25 frame and the current frame is implemented. Such correlation search allows estimating a
delay parameter and then a gain. The gain corresponds to the amplitude ratio between the
filtered part of the current frame and the signal predicted by applying the delay.
For example, 6 bits may be allocated for the gain and 6 bits for the delay. The third
number of bits budg3 is then equal to 12.
30 The num_bits variable may then be updated:
num_bits := num_bits – budg3.
The second number of bits budg2 is then equal to the updated num_bits variable.
20
Figure 5 illustrates an audio decoder 500 according to an embodiment of the
invention and Figure 6 is a diagram illustrating the steps of a method of decoding
according to an embodiment of the invention, implemented in the audio decoder 500 of
Figure 5.
The decoder 500 comprises a receiving unit 501 for receiving, 5 at step 601, the
coded digital signal (or bit flow) originating from the coder 500 of Figure 1. The bit flow is
submitted to a categorising unit 502 able to determine, at step 602 if the current frame is a
CELP frame, an MDCT frame, or a transition frame. For this purpose, the categorising
units 502 is able to deduct a bit flow from the bit flow information indicating whether or
10 not the current frame is a transition frame, and an information indicating which CELP core
to use for decoding a CELP frame or a transition CELP sub-frame.
At step 603, it is verified that the current frame is a transition frame.
If the current frame is not a transition frame, it is verified at step 604 that the
current frame is a CELP frame. If that is the case, the frame is transmitted to a CELP
15 decoder 504 able to decode a CELP frame at step 605, with the core frequency indicated
by the categorising units 502. After decoding a CELP frame, the CELP decoder 504 may
store, at step 606, in a memory 506, parameters such as the linear prediction filter
coefficients A(z) and internal states such as predictive energy in cases where the following
frame is a transition frame.
20 As an output from the CELP decoder 504, the signal may be re-sampled, at step
607, with the output frequency of the decoder 500 by a re-sampling unit 505. In an
embodiment of the invention, the re-sampling unit comprises an FIR filter and re-sampling
introduces a delay of (for example) 1.25 ms. In an embodiment, post-processing may be
applied to CELP decoding before or after re-sampling.
25 As mentioned above, in an embodiment, extending the band may also be
performed, by a managing unit of band extension 5051 at steps 6071 and 6151, with
decoding associated with the high band when the current frame is of CELP type. The high
band is then combined to CELP coding with potentially an additional delay applied to the
CELP synthesis at low band.
30 The signal decoded by the CELP decoder and re-sampled, potentially postprocessed
before or after re-sampling, is transmitted to an output interface 510 of the
decoder at step 608.
The decoder 500 further comprises an MDCT decoder 507. In cases where has been
determined at step 604 that the current frame is an MDCT frame, the MDCT decoder 507
21
is able to decode the MDCT frame in a classic manner at step 609. Furthermore, a delay
corresponding to the delay necessary for the re-sampling application of the signal
originating from the CELP decoder 504 is applied at the decoder output by a delay unit
508, so as to synchronise the MDCT synthesis with the CELP synthesis, at step 610. The
signal decoded by MDCT and delayed is transmitted to the output 5 interface 510 of the
decoder at step 608.
In cases where the current frame is determined as being a transition frame after step
603, a device for determining the distribution of bits 503 is able to determine, at step 611,
the first number of bits budg1 allocated for CELP coding the transition frame and the
10 second number of bits budg3, allocated for transform coding the transition frame. The
device 503 may correspond to the device 700 described in details in reference to Figure 7.
The MDCT decoder 507 uses the third number of bits budg3 calculated by the
determining unit 503 for adjusting the rate necessary for decoding the transition frame. The
MDCT decoder 507 further zeroes the memory of the MDCT transformation and decodes
15 the transition frame at step 612. The signal originating from the MDCT decoder is then
delayed by the delay unit 508 at step 613.
In parallel, the CELP decoder 504 decodes the transition CELP sub-frame based on
the first number of bits budg1, at step 614. For this purpose, the CELP decoder 504
decodes the CELP parameters that may depend on the current frame category, and
20 comprising for example pitch values from the adaptive dictionary, the fixed and gains
dictionary of the CELP sub-frame, and uses the linear prediction filter coefficients.
Furthermore, the CELP decoder 504 updates the CELP decoding states. These states may
typically comprise the predictive energy of the innovation originating from the preceding
CELP frame for generating the signal sub-frame over 4 ms or 5 ms according to whether
25 the 12.8 kHz or 16 kHz CELP core is being used (in the case of restricted coding the
transition CELP sub-frame).
As previously mentioned, the application WO2012/085451 provides extra coding a
sub-frame of 5 ms for the 12.8 kHz CELP core and two extra sub-frames of 4 ms for the 16
kHz CELP core.
30 As explained in reference to Figure 3, in the 12.8 kHz case, the 6.25 ms delay is not
filled and the overlap-add is affected: the decoder only has 0.625 ms of overlap-add, which
is insufficient.
22
In the 16 kHz case, to extra CELP sub-frames are coded at the start of the transition
frame, which only leaves very little budget for coding the transition MDCT frame and may
lead to a quality decrease with respect to MDCT coding at “full rate” in the current frame.
Thus, the solution of the international application WO2012/085451 is not
5 satisfying.
An independent aspect of the invention provides, from a single extra transition
CELP sub-frame, partially generating a second sub-frame by reusing the coding parameters
used for coding the transition CELP sub-frame. The delay is thus filled, by ensuring
sufficient overlap-add, and without affecting the MDCT coding rate of the transition frame.
10 For this purpose, the invention also aims for a method of decoding a coded digital
signal P, in a decoder 500 able to decode the signal frames according to predictive
decoding or according to transform decoding, comprising the following steps:
receiving,
at step 501, a first set of predictive coding parameters coding a first
digital signal frame;
15 predictive
decoding, at step 605, the first frame based on the first set of predictive
coding parameters;
receiving,
at step 501 for a new frame, a second set of parameters for predictive
coding a first transition sub-frame of a transform coded transition frame,
decoding,
at step 614, the first transition sub-frame based on the second set of
20 predictive coding parameters,
generating,
at step 614, samples from the second transition sub-frame, from at least
one predictive coding parameter of the second set.
The invention further aims for the decoder 500 for implementing the method of
decoding P, as well as a computer program comprising the instructions for performing the
25 method of decoding P, when these instructions are executed by a processor.
The CELP parameters reused for generating the second sub-frame may be the gain
vector, the adaptive dictionary vector, and the fixed dictionary vector.
According to an embodiment of the method of decoding P, a minimum overlap value
may be predefined for transform decoding and the number of generated samples from the
30 second sub-frame is determined based on the minimum overlap value. This last sub-frame
may be generated without extra information by prolonging the CELP synthesis by
repeating the pitch prediction with the same pitch delay and the same adaptive dictionary
gain as in the first sub-frame, and by performing a synthesis LPC filtering with the same
LPC coefficients and a de-emphasis or de-accentuation.
23
The second CELP sub-frame may then be truncated so as to only preserve 1.25 ms of
signal in the case of the 12.8 kHz CELP core, and 2.25 ms of signal in the case of the 16
kHz CELP core. The first CELP sub-frame is thus completed so as to have 6.25 ms of
extra signal allowing filling the gap and ensuring a satisfying overlap-add (minimum
overlap value, for example of 1.875 ms) with the MDCT transition 5 frame. In an
embodiment, the extra CELP sub-frame has a length extended to 6.25 ms for the 12.8 and
16 kHz CELP cores, which implies modifying “normal” CELP coding for having such
length of extended sub-frame, in particular for the fixed dictionary.
In addition to the previous embodiment of the method of decoding P, the method P
10 may further comprise a step 615 of re-sampling performed by a finite impulse response
filter. As previously explained, the FIR filter may be integrated into the re-sampling unit
505. Re-sampling uses the FIR filter memory from the preceding CELP frame and
processing induces an extra delay of 1.25 ms in this example.
The method P may further involve a step of adding an additional signal obtained from
15 samples stored in the finite impulse response filter memory, to fill the delay introduced by
the re-sampling step. Thus, 1.25 ms of signal, in addition to the 6.25 ms of additional
signal previously generated, are generated by the decoder 500, these samples
advantageously allowing filling the delay introduced by the re-sampling of 6.25 ms of
additional signal.
20 For this purpose, the FIR filter memory of the re-sampling unit 505 may be saved for
each frame after CELP decoding. The number of samples in this memory corresponds to
1.25 ms at the CELP core frequency considered (12.8 or 16 kHz).
According to a complementary embodiment of the method P, re-sampling the stored
samples is performed by an interpolation method introducing a second delay shorter than
25 the first delay from the finite impulse response filter, which may be considered as null.
Thus, the 1.25 ms of signal generated from the FIR filter memory, are re-sampled
according to a method implying a minimum delay. For example, re-sampling the 1.25 ms
of signals generated by the FIR filter memory may be performed by cubic interpolation,
which implies a delay from two samples only, a minimum delay compared to the delay
30 from the FIR filter. Thus, the two extra signal samples are required for re-sampling the
above-mentioned 1.25 ms of signal: these two extra samples may be obtained by repeating
the last value of re-sampling memory of the FIR filter.
The decoder may further decode the high-frequency part from the 6.25 ms of CELP
signal obtained from the first and second transition frames. For this purpose, the CELP
24
decoder 504 may use the adaptive gain and the fixed dictionary vector from the last subframe
of the preceding CELP frame.
The decoder 500 further comprises an overlap-add unit 509 able to ensure the overlapadd,
at a step 616, between the decoded and re-sampled CELP transition sub-frames, the
samples re-sampled by cubic interpolation, and the decoded signal of 5 the transition frame
originating from the MDCT decoder 507.
For this purpose, the unit 509 applies the synthesis modified window 327 of Figure 3.
Thus, prior to the MDCT aliasing point for the two first quarters, the samples are zeroed.
After the above-mentioned aliasing point, the windowed samples are divided by the non10
modified window 324 of Figure 3, and multiplied by a sinus-type window so that,
combined with the window applied to the encoder, the total window is sin². In the part
concerned by the overlap-add, the samples originating from the CELP and 0-delay resampling
(for example by cubic interpolation) are weighted by a cos² window.
The transition frame thus obtained is transmitted to the output interface 510 of the
15 decoder at step 608.
Figure 7 represents an example of device 700 for determining the distribution of bits
for a transition frame.
The device comprises a random access memory 704 and a processor 703 for storing
instructions allowing implementing the method of determining the distribution of bits for a
20 transition frame described above. The device also involves a mass memory 705 for storing
data intended for being preserved after applying the method. The device 700 further
involves an input interface 701 and an output interface 706 intended for receiving the
digital signal frames and for emitting the detail for the budget allocated to these different
frames, respectively.
25 The device 700 may further involve a digital signal processor (DSP) 702. This DSP
702 receives digital signal frames for forming, demodulating, and amplifying these frames,
in a known manner known per se.
The present invention does not limit itself to the embodiments described above for
example purposes; it extends to other variants.
30 Thus, an embodiment has been described wherein the compression or decompression
devices are entities as a whole. Of course, the devices may be embedded in all types of
25
more significant devices such as for example a digital camera, a photo camera, a mobile
phone, a computer, a cinema projector, etc.
Moreover, an embodiment proposing a particular design of the compression,
decompression, and comparison devices has been described. These designs are only given
for illustrative purposes. Thus, an arrangement of the components 5 and a different
distribution of the tasks assigned for each of the components may also be considered. For
example, the tasks performed by the digital signal processor (DSP) may also be performed
by a classic processor.
10
15
26
CLAIMS
1. A method of determining the distribution of bits for coding a transition frame (321;
322), said method being implemented in a coder/decoder (100; 500) for
coding/decoding a digital signal, the transition frame being preceded by a
predictive coded preceding frame (320), coding the transition 5 frame comprising
transform coding and predictive coding a single sub-frame of the transition frame,
the method comprising the following steps:
assigning
(402; 405) a bit rate for predictive coding of the transition sub-frame, said
bit rate being equal to the minimum between the bit rate for transform coding the
10 transition frame and a first predetermined bit rate value;
determining
(404; 408) whether the first number of bits allocated for predictive
coding the transition sub-frame for said bit rate; and
calculating
(410) a second number of bits allocated for transform coding the
transition frame from the first number of allocated bits and a number of bits available
15 for coding the transition frame.
2. The method according to claim 1, wherein the coder/decoder (100; 500) comprises
a first core operating, for predictive coding/decoding a signal frame, at a first
frequency, and a second core operating, for predictive coding/decoding a signal
frame, at a second frequency,
20 wherein the first predetermined bit rate value of depends on the core selected from
the first and second cores for coding/decoding the predictive coded preceding
frame (320).
3. The method according to claim 2, when the first core has been selected for
coding/decoding the predictive coded preceding frame (320), the assigned bit rate is
25 further equal to the maximum between the bit rate for the transform coded
transition frame (322) and at least one second predetermined bit rate value, the
second value being lower than the first value.
4. The method according to one of the previous claims, wherein the digital signal is
decomposed into at least one frequency low band and one frequency high band,
30 wherein the first calculated number of bits is assigned for predictive coding the
transition sub-frame (321) for the frequency low band, and wherein a third
27
predetermined number of bits is allocated for coding the transition sub-frame for
the frequency high band,
and wherein the second number of bits allocated for transform coding the transition
frame (322) is further determined from the third predetermined number of bits.
5. The method according to claim 4, wherein the number of bits 5 available for coding
the transition frame (321; 322) is fixed.
6. The method according to claim 5, wherein the second number of bits is equal to the
fixed number of bits for coding the transition frame (321; 322) minus the first
number of bits minus as the third number of bits.
10 7. The method according to claim 5, wherein the second number of bits is equal to the
fixed number of bits for coding the transition frame (321; 322) minus the first
number of bits minus the third number of bits minus a first bit minus a second bit,
the first bit indicating whether low-pass filtering is performed when determining
the predictive coding parameters of the transition sub-frame, said parameters being
15 related to the tonal lead time,
the second bit indicating the frequency used by the coder/decoder core for
predictive coding/decoding the transition sub-frame.
8. A method of coding a digital signal in a coder (100) able to code signal frames
according to predictive coding or according to transform coding, comprising the
20 following steps:
 coding a preceding frame (301) of digital signal samples according to predictive
coding;
 coding a current frame (302) of digital signal samples into a transition frame
(321; 322): coding the transition frame (321; 322) comprising transform coding
25 and predictive coding a single sub-frame (321) of the transition frame, said
coding of the current frame (302) comprising the following sub-steps:
determining
(209) the distribution of bits according to one of the
previous claims;
transform
coding (212) the transition frame (322) on the second
30 number of allocated bits;
28
predictive
coding (213) the transition sub-frame (321) and the first
number of allocated bits.
9. The method of coding according to claim 8, wherein predictive coding comprises
generating determined predictive coding parameters for said allocated bit rate.
10. The method of coding according to one of claims 8 and 5 9, wherein predictive
coding comprises generating predictive coding parameters restricted with respect to
predictive coding the preceding frame (320) reusing at least one parameter for
predictive coding of the preceding frame.
11. A method of decoding a coded digital signal, implemented in a decoder (500) able
10 to decode signal frames according to predictive coding or according to transform
decoding, comprising the steps of:
 predictive decoding (605) a preceding frame of digital signal samples coded
according to predictive coding;
 decoding a transition frame (321; 322) coding a current frame of digital
15 signal samples, coding the transition frame comprising transform coding
and predictive coding a single sub-frame (321) of the transition frame,
comprising the sub steps of:
- determining (611) the distribution of bits according to one of Claims 1
to 7;
20 - predictive decoding (614) the transition sub-frame (321) on the first
number of allocated bits;
- transform decoding (612) the transition frame (322) on the second
number of allocated bits.
12. A computer program comprising instructions for implementing the method
25 according to any of the previous claims, when these instructions are executed by
processor.
13. A device for determining a distribution of bits for coding a transition frame (321;
322), said device (104; 503) being implemented in a coder/decoder for
coding/decoding a digital signal, the transition frame being preceded by a
30 predictive coded preceding frame (320), coding the transition frame comprising
transform coding and predictive coding a single sub-frame (321) of the transition
29
frame, the number of bits for coding the transition frame being fixed, the device
comprising a processor arranged for performing the following operations:
- assigning a bit rate for predictive coding the transition sub-frame, said
bit rate being equal to the minimum between the bit rate for transform
coding the transition frame and a first pre-determined 5 bit rate value;
- determining a first number of bits allocated for predictive coding the
transition sub-frame for said bit rate;
- calculating a second number of bits allocated for transform coding the
transition frame from the number of bits required for coding the coding
10 parameters and the fixed number of bits for coding the transition frame.
14. A coder able to code digital signal frames according to predictive coding or
transform coding, comprising :
 a device (104) according to claim 13;
 a predictive coder (103) comprising a processor arranged for performing the
15 following operations:
coding
a preceding frame of digital signal samples according to predictive
coding;
predictive
coding a single sub-frame comprised in a transition frame coding
a current frame of digital signal samples, coding the transition frame
20 comprising transform coding and predictive coding said sub-frame, the
processor being arranged for predictive coding the transition sub-frame on
the first number of allocated bits;
a
transform coder (105) comprising a processor arranged for performing the
operation of transform coding the transition frame on the second number of
25 allocated bits.
15. A decoder for a digital signal coded by predictive coding and by transform coding,
comprising
 a device (503) according to claim 13;
 a predictive decoder (504) comprising a processor arranged for performing the
30 following operations:
predictive
decoding a preceding frame (320) of digital signal samples
coded according to predictive coding;
30
predictive
decoding a single sub-frame (321) comprised in a transition
frame coding a current frame of digital signal samples, coding the
transition frame comprising transform coding and predictive coding said
sub-frame, the processor being arranged for performing the operation of
predictive decoding the transition sub-frame on 5 the first number of
allocated bits;
 a transform decoder (507) comprising a processor arranged for performing the
operation of transform decoding the transition frame (222) on the second number of
allocated bits.