Advanced processing based on a complex-exponential-modulated filterbank

DESCRIPTION

Advanced processing based on an exponentially complex modulated filter bank and adaptive time signaling procedures 5

TECHNICAL FIELD

[0001] The present invention relates to audio source coding systems, but the same

procedures could also be applied in many other technical fields. Different techniques are introduced that are useful for audio coding systems that use parametric representations of stereo properties.

BACKGROUND OF THE INVENTION AND PREVIOUS TECHNIQUE

fifteen [0002] The present invention relates to the parametric coding of the stereo image of a signal of

Audio. The usual parameters used to describe the properties of stereo images are the inter-channel intensity difference (IID), the inter-channel time difference (ITD) and the inter-channel coherence (IC , inter-channel coherence). To reconstruct the stereo image based on these parameters, a procedure is required that can reconstruct the correct level of correlation between the two channels, according to the IC parameter. This is achieved by a decorrelation procedure.

[0003] There are a couple of procedures available for creating decorrelated signals. Ideally, a time and linear invariant function (LTI) with a response of

25 total step frequency An obvious procedure to achieve this is to use a constant delay. However, using a delay, or any other LTI total pass functions, will result in a total no pass response after adding the unprocessed signal. In the case of a delay, the result will be a comb filter. The comb filter frequently provides an undesirable "metallic" sound that, even if the stereo widening effect can be efficient, greatly reduces the naturalness of the original.

30

[0004] Methods in the frequency domain for generating a decorrelated signal are also known from the prior art by adding a random signal to the IID values ​​along the frequency axis, where different sequences are used for the different audio channels. A problem with the decorrelation of the frequency domain by random sequence modifications is the introduction of pre-echoes. Subjective tests

35 they have shown that for non-stationary signals, pre-echoes are much more annoying than post-echoes, which is also well supported by established psychoacoustic principles. The problem could be reduced by dynamically adapting transform sizes to the signal characteristics, in terms of transient content. However, switching transform sizes is always a difficult (i.e., binary) decision that affects the entire bandwidth of the signal and can be difficult to achieve consistently.

40

[0005] US Patent Application Publication US 2003/0219130 A1 describes a coherence-based audio synthesis and coding. In particular, an auditory scene is synthesized from a mono audio signal by modifying, for each critical band, a parameter of the auditory scene, such as an interaural level difference (ILD) and / or a difference of interaural time (ITD) for

Four. Five each sub-band within the critical band, where the modification is based on an estimated average consistency for said critical band. Consistency-based modification produces auditory scenes with object widths, which reproduce much more accurately the widths of the objects in the original input auditory scene. Stereo parameters are well-known BCC parameters, in which BCC stands for biaural signal coding. When two different decorrelated output channels are generated, the 50 frequency coefficients that are obtained by a discrete Fourier transform are grouped together into a single critical band. Based on the measurement of coherence between channels, the weighting factors are multiplied by a pseudorandom sequence, which is preferably chosen so that the variance is approximately constant for all critical bands, and the average is "0" within each band review. The same sequence applies to the spectral coefficients of each different frame.

55

[0006] WO 91/20167 describes a method and apparatus for creating decorrelated audio output signals and audio recordings made in this way. In one embodiment, the reference describes the use of bandpass filters and band shift circuits. In particular, a plurality of band 1 through M pass filters are provided to decompose an input signal into band pass filter output signals.

60 Each bandpass filter output signal is introduced into a phase shift network. It is considered a

band of critical bandwidth centered on a frequency. If the frequency bands used are much smaller than the critical bandwidth, then the critical frequency band in question will be formed by a plurality of subbands, each with a different phase shift. The critical band in question will have an average phase shift that is an average of the individual phase shifts. If the 5 signal source consists of an analog signal, it can be converted to digital format through an analog-digital converter. To apply the filter functions, an additional embodiment has a conversion operation with a filter function. The input signal is convolved with different filter functions, to obtain once again a certain average value. US patent US6005946 describes a method of decorrelation that applies frequency band reverberation filtering.

10

DESCRIPTION OF THE INVENTION

[0007] The objective of the present invention is to provide an apparatus and a method for generating a decorrelation signal.

15

[0008] This objective is achieved by an apparatus for generating a decorrelation signal according to claim 1, a method for generating a decorrelation signal according to claim 13, a multichannel decoding method according to claim 14 or a computer program according to claim 15.

20

[0009] The present invention is based on the finding that, on the decoding side, a good decorrelation signal is obtained to generate a first and a second channel of a multichannel signal based on the mono input signal, when It uses a reverberation filter, which introduces an entire or preferably fractional delay to the input signal. It is important that this reverberation filter is not applied to the entire signal of

25 entry. On the contrary, several reverberation filters are applied to several subbands of the original input signal, that is, the mono signal, so that the reverberation filtering used by the reverberation filters is not applied in the time domain or in the frequency domain, that is, in the domain that is reached when a Fourier transform is applied. In an inventive manner, the reverberation filtering using reverberation filters for the subbands is carried out individually in the subband domain.

30

[0010] A sub-band signal includes a sequence of at least two sub-band samples, the sub-band samples representing a bandwidth of the sub-band signal, which is less than the bandwidth of the input signal Of course, the frequency bandwidth of a sub-band signal is greater than the frequency bandwidth attributed to a frequency coefficient obtained by transforming

35 of Fourier. Sub-band signals are preferably generated by a filter bank having for example 32 or 64 channels of the filter bank, while an FFT would have, for the same example, 1024 or 2048 frequency coefficients, that is, channels of frequency.

[0011] Subband signals may be subband signals that are obtained by filtering 40 subband of a sample block of the input signal. Alternatively, the subband filter bank

It can also be applied continuously without block processing. However, for the present invention, block processing is preferred.

[0012] Since reverberation filtering is not applied to the entire signal, but is applied by sub-bands, a "metallic" sound caused by comb filtering is avoided.

[0013] In cases where the sample period between two consecutive subband samples of the subband is too large for a good sound impression at the end of the decoder, it is preferable to use fractional delays in a reverb filter, such as a delay between 0.1 and 0.9, and preferably 0.2 and 0.8,

fifty of the sampling period of the subband signal. It should be noted that, in the case of critical sampling, and when 64 sub-band signals are generated using a filter bank that has 64 channels of the filter bank, the sampling period in a sub-band signal is 64 times greater than The sampling period of the original input signal.

55 [0014] It should be noted that delays are an integral part of the filtering process used in the

reverberation device. The output signal consists of multiple delayed versions of the input signal. It is preferable to delay signals in fractions of the sub-band sampling period, in order to achieve a good reverberation device in the sub-band domain.

[0015] In preferred embodiments of the present invention, the delay, and preferably the fractional delay introduced by each reverberation filter in each subband, is the same for all subbands. However, the filter coefficients are different for each of the subbands. It is preferable to use IIR filters. Depending on the actual situation, the fractional delay and the filter coefficients for the different filters may

5 be determined empirically using hearing tests.

[0016] The subbands filtered by the set of reverberation filters constitute a decorrelation signal that has to be mixed with the original input signal, ie the mono signal, to obtain a decoded left channel and a decoded right channel. This mixture of a decorrelation signal with the signal

10 The original is carried out based on a parameter of coherence between channels, transmitted together with the parameterically encoded signal. To obtain different left and right channels, that is to say, first and second different channels, the mixing of the decorrelation signal with a mono signal to obtain the first output channel is different from the mixture of said decorrelation signal with the mono signal for Get the second output channel.

15

[0017] For greater efficiency on the coding side, multichannel coding is carried out using an adaptive determination of the stereo parameter set. In this regard, the encoder includes, in addition to means for calculating the mono signal and in addition to means for generating a set of stereo parameters, means for determining a validity of the sets of stereo parameters for subsequent parts of

twenty the left and right channels. Preferably, the determination means are operative to activate the generation means, when it is determined that the set of stereo parameters ceases to be valid, so that a second set of stereo parameters is calculated for parts of the left and right channels that begin In a second time limit. This second time limit is also determined by means of determining validity.

25

[0018] The encoded output signal then includes the mono signal, a first set of stereo parameters and a first time limit associated with the first set of parameters and the second set of stereo parameters, and the second time limit associated with The second set of stereo parameters. On the decoding side, the decoder will use a valid stereo parameter set until it is reached

30 A new time limit. When this new time limit is reached, decoding operations are carried out using the new set of stereo parameters.

[0019] Compared to prior art procedures, which performed block processing and, therefore, a block determination of stereo parameter sets, the determination

35 Adaptive adaptive stereo parameter sets for different time limits determined on the encoder side provide on the one hand a high coding efficiency, and on the other hand a high coding quality. This is due to the fact that for relatively stationary signals, the same set of stereo parameters can be used for many blocks of the mono signal samples, without introducing audible errors. On the other hand, when non-stationary signals are involved, the determination of stereo parameters

40 Adaptive adaptive provides improved temporal resolution, so that each signal part has its optimum stereo parameter set.

[0020] The present invention describes a solution to prior art problems using a

reverberation unit as a decorator implemented with fractional delay lines in a bank of 45 filters, and using an adaptive level adjustment of the re-related reverberated signal.

In the following, several aspects of the present invention are briefly explained. An aspect of the invention is a method for modifying complex value subband signals by filtering each complex value subband signal with a finite impulse response filter, where the finite impulse response filter for the number of sub-band n is given by a Fourier transform of the form

íexp (-ix (n +1 / 2) r) Gr (ú)), for n pair;

Hn (oj) - <,

[exp (- / Vr (n + l / 2) r) Gf (o? +; r), for odd n.

fifty where the parameter t = T / L, and where the bank of

Synthesis filters have L subbands and the desired delay is T measured in sample units of the output signal.

Another aspect of the invention is a method for modifying complex value subband signals by filtering, where the Gr (w) filter satisfies approximately Vz (w) Gz (w) + Vz (w + n) Gz (w + n) = l, where Vz (w) is the

vr (^) = A ik ^ p (l) p (l -T - Lk),

discrete temporal Fourier transform of the sequence ', and p (l) is the filter

prototype of said complex filter bank and A is the appropriate real normalization factor.

Another aspect of the invention is a method for modifying the subband signals of complex value 5 by filtering, where the filter Gr (m) satisfies Gr (-a)) = Gr (a) + n) *, thereby that the even index impulse response samples are of real value and the odd index impulse response samples are of pure imaginary value.

[0021] Another aspect of the invention is a method for generating stereo decorrelation for parametric stereo reconstruction, by means of a decoder, applying an artificial reverberation process

to synthesize the lateral signal.

Another aspect of the invention is a method for generating stereo decorrelation for parametric stereo reconstruction, by means of a decoder, performing the reverberation process within a complex modulated filter bank using phase delay adjustment in each channel of the filter bank.

15

BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Next, the present invention will be described by illustrative examples, without limiting the scope of the invention, with reference to the accompanying drawings, in which:

20

Figure 1 shows a block diagram of the inventive apparatus;

Figure 2 shows a block diagram of the means for generating a correlated signal;

25 Figure 3 shows the analysis of a single channel and the synthesis of the pair of stereo channels based on the reconstructed stereo subband signals, in accordance with the present invention;

Figure 4 shows a block diagram of the division of the parametric stereo parameter sets into time segments, based on the signal characteristics; and 30

Figure 5 shows an example of the division of the parametric stereo parameter sets into time segments, based on the signal characteristic.

DESCRIPTION OF PREFERRED EMBODIMENTS

35

[0023] The embodiments described below are merely illustrative of the principles of the present invention for parametric stereo coding. It should be understood that modifications and variations of the provisions and details described herein will be apparent to those skilled in the art. Therefore, the intention is to be limited only by the scope of the independent patent claims and

40 not through the specific details presented by way of description and explanation of the embodiments of this document.

[0024] Delaying a signal by a fraction of a sample can be achieved by various interpolation procedures of the prior art. However, special cases arise when the signal

Four. Five Original is available as oversampled complex value samples. Performing the fractional delay in the qmf bank by applying only phase delay using a factor for each qmf channel, corresponding to a constant time delay, results in several artifacts.

[0025] This can be effectively avoided by using a compensation filter in accordance with a novel approach 50 that allows high quality approximations to arbitrary delays in any complex filter bank.

exponential-modulated. A detailed description follows.

A continuous time model

55 [0026] To simplify the calculations, a filter bank of the complex exponential modulated L-band filter is

In this case, you will model using a continuous time window transform using the synthesis waveforms.

where n, k are integers with n> 0, and Q is a fixed phase term. Results are obtained for discrete time signals by proper sampling of variable t with 1 / L separation. It is assumed that the real value window 5 v (t) is chosen so that for real value signals x (t) it is very precise that

yes

"C

cítC¿) = (3)

—EQ

10

where * indicates complex conjugation. It is also assumed that v (t) is essentially limited in band to the frequency range [-n, ñ]. Consider the modification of each frequency band n by filtering the discrete temporal analysis samples cn (k) with a pulse response filter hn (k),

(4)

15 >

So, the modified synthesis

20

can be calculated in the frequency domain to be

(6)

j> (o ») = H (o)) x {<ü),

25 Where f (w) indicates Fourier transforms of f (t) and

[0027] Here, Hn (m) = Skh „(k) exp (-ikrn) is the discrete temporal Fourier transform of the filter applied in the

30 frequency band n for n> 0 and

Hn {(ú) = // _, _ „(- <») * for n <0. (8)

Here, it should be noted that the special case Hn (m) = 1 leads to H (w) = 1 in (7), due to the special design of the window in v (t). Another case of interest is Hn (rn) = exp (-iw), which provides H (w) = exp (-iw), so that y (t) = x (t- 1).

The proposed solution

40 [0028] To obtain a delay of size t, such as y (t) = x (tt), the problem is to design Hn (m) filters to

n> 0, so that

where H (w) is given by (7) and (8). The particular solution proposed in this case is to apply the filters

5

10

Here, GT (-m) = GT (m + n) * implies consistency with (8) for all n. The introduction of (10) on the right side of (7) results in

where VT (a>) = Znb ((on (2n + l / 2)), with b {co) -exp {¡rco) \ v {co) \ 2 elementary calculations show that VT (a>) is the transformed discrete temporary Fourier of 15

Very good approximations to the perfect delay can be obtained by solving the linear system

20

in the sense of least squares with a filter FIR filter coefficients, equation (13) can be written

In terms of

2Xv, (2 * - /) g, (0 = á [t], (14)

25

where 5 [k] = 1 for k = 0 and 5 [k] = 0 for k 4 0.

[0029] In the case of a discrete temporary L-band filter bank with prototype filter p (k), the delay 30 obtained in sample units is Lty the calculation (12) is replaced with

vr (k) = ikYjp (l) p (lT-Lk), (15)

where T is the integer closest to Lt. In this case, p (k) extends with zeros outside its support. For a finite length prototype filter 35, there is only a finite number of vr (k) other than zero, and (14) is a system of linear equations. The unknown number of g ^^ (k) is typically chosen to be a small number. For good designs of QMF filter banks, 3-4 shots already provide a very good delay behavior. Besides, the

dependence of the filter sockets gt (k) with the delay parameter t can often be modeled satisfactorily by low order polynomials.

Adaptive time grid signaling for stereo parameters 5

[0030] Parametric stereo systems always lead to compromises in terms of limited temporal or frequency resolution to minimize transported data. However, it is well known by psychoacoustics that some spatial signals may be more important than others, which leads to the possibility of ruling out less important signals. Therefore, the temporary resolution does not have to be

10 constant. A large gain in bit rate can be obtained by allowing the temporary grid to synchronize with spatial signals. This can easily be done by sending a variable number of parameter sets for each data frame corresponding to a fixed-sized time segment. To synchronize the parameter sets with corresponding spatial signals, it is necessary to send additional time grid data describing the temporal position for each parameter set. The resolution of these 15 temporary pointers could be chosen to be very low to keep the total amount of data minimized. A special case where a time signal for a set of parameters coincides with the beginning of a frame, could be explicitly signaled to avoid sending said time pointer.

[0031] Figure 4 shows an inventive apparatus for carrying out parameter analysis for segments of time that have variable time limits and dependent on the signal. The inventive apparatus includes means 401

to divide the input signal into one or more time segments. The time limits separating the time segments are provided by means 402. The means 402 use a detector specially designed to extract spatial signals that are relevant to decide where to place the time limits. Media 401 emits all input signals divided into one or more time segments. This output is introduced in means 403 for independent parametric analysis for each time segment. The means 403 deliver a set of parameters for each time segment analyzed.

[0032] Figure 5 shows an example of how the time grid generator can function for a hypothetical input signal. In this example, a set of parameters is used for each data frame, if not

30 No other time limit information is present. Therefore, when there is no other time limit information, the inherent time limits of the data frame are used. The time limits represented in Figure 5 are emitted from the means 402 of Figure 4. The time segments represented in Figure 5 are provided by means 401 of Figure 4.

35 [0033] The apparatus for encoding a stereo signal in order to obtain a mono output signal and the

set of stereo parameters, includes the means to calculate the mono signal by combining a left and right channel of the stereo signal, by means of a weighted sum. Additionally, means 403 generate a first set of stereo parameters using a part of the left channel and a part of the right channel, said parts starting at a first time limit, being connected to the means to determine the validity of the first set of parameters stereo for subsequent parts of the left channel and the right channel.

[0034] The determination means are collectively formed by means 402 and 401 of the

Figure 1.

Four. Five [0035] In particular, the determination means are operative to generate a second time limit and

to activate the generation means, when it is determined that this first set of stereo parameters is no longer valid, so that a second set of stereo parameters is generated for parts of the left and right channels that begin at the second time limit.

fifty [0036] In Figure 4 no means are shown for delivering the mono signal, the first set of parameters

stereo and the first time limit associated with the first set of stereo parameters, and the second set

of stereo parameters and the second time limit associated with the second set of stereo parameters, such as the stereo signal coded parametrically. The means for determining the validity of a set of stereo parameters may include a transient detector, since there is a high probability that, after a transient, it is necessary to generate a new stereo parameter, since the signal has significantly changed its profile. Alternatively, the means of determining validity may include a synthesis analysis device, which is adapted to decode the mono signal and the set of stereo parameters in order to obtain a decoded left channel and a decoded right channel, in order to compare the left decoded channel and the decoded right channel with the left channel and the channel

right, and to activate the generation means, when the left decoded channel and the decoded right channel differ from the left channel and the right channel by more than a predetermined threshold.

[0037] Data frame 1: the time segment corresponding to parameter set 1 begins at the beginning of data frame 1 since no other time limit information is present in this

data frame.

[0038] Data frame 2: Two time limits are present in this data frame. The time segment corresponding to parameter set 2 begins at the first time limit in this data frame.

10 The time segment corresponding to parameter set 3 begins at the second time limit in this data frame.

[0039] Data frame 3: a time limit is present in this data frame. The time segment corresponding to parameter set 4 begins at the time limit in this data frame.

15

[0040] Data frame 4: a time limit is present in this data frame. This time limit coincides with the initial limit of the data frame 4 and does not need to be signaled since it is contemplated in the default case. Therefore, this time limit can be removed. The time segment corresponding to parameter set 5 starts at the beginning of data frame 4, even without signaling this time limit.

20

Use of artificial reverberation as a decorrelation procedure for parametric stereo reconstruction

[0041] A fundamental part of performing stereo synthesis in a parametric stereo system is to reduce the coherence between the left and right channels to create amplitude of the stereo image. This can

25 be made by adding a filtered version of the original mono signal to the lateral signal, where the lateral and mono signals are defined by:

mono = (left + right) / 2, and

30 lateral = (left - right) / 2, respectively.

[0042] In order not to change the bell too much, the filter in question should preferably be of a total pitch. A satisfactory approach is to use similar full-pass filters used for reverberation processes

artificial. Artificial reverberation algorithms normally require a high temporal resolution to provide a pulse response that is satisfactorily diffused over time. There are great advantages in basing an artificial reverberation algorithm on a complex filter bank, such as the complex qmf bank. The filter bank provides excellent possibilities to make reverberation properties selective in frequency, in terms of, for example, equalization of reverberation, fall time, density and timbre. However, filter bank implementations normally exchange temporary resolution for higher frequency resolution, which usually complicates implementing a reverberation process that is smooth enough over time. To address this problem, a novel procedure should use a fractional delay approximation, applying only one phase delay by a factor for each channel 45 qmf corresponding to a constant time delay. This primitive fractional delay procedure introduces temporary persistence, which fortunately is very convenient in this case. Such temporal persistence contributes to temporal diffusion, which is very desirable for reverberation algorithms and increases as the phase delay approaches pi / 2 or - pi / 2.

fifty [0043] For natural reasons, artificial reverberation processes are processes with a finite response to

momentum, and offer natural exponential drops. In document [PCT / SE02 / 01372] it is pointed out that if a reverberation unit is used to generate a stereo signal, the reverberation drop may sometimes be unwanted after the end of a sound. However, these unwanted reverberation tails can easily be attenuated or completely eliminated simply by varying the gain of the reverberation signal. 55 A detector designed to find sound endings can be used for this purpose. If the reverberation unit generates artifacts in some specific signals, for example, transients, a detector can also be used for said signals to attenuate them.

[0044] Figure 1 shows an inventive apparatus for the signal decorrelation procedure that is

60 used in a parametric stereo system. The inventive apparatus includes means 101 for providing a series of

subband signals. The provisioning means may be a complex QMF filter bank, where each signal is associated with a sub-band index.

[0045] Sub-band signals emitted by means 101 of Figure 1 are introduced into means 5 102 to provide a decorrelated signal 102, and into means 103 and 106 to modify the subband signal. The output of 102 is introduced in means 104 and 105 to modify the signal, and the outputs of 103, 104, 105 and 106 are introduced in means 107 and 108 to sum the sub-band signals.

[0046] In the embodiment of the presently described invention, the modification means 103, 104, 105 and 106 of the sub-band signals regulate the level of the decorrelated signal and the unprocessed signal that is the output

of 101, multiplying the sub-band signal by a gain factor, so that each sum of each pair results in a signal with the amount of decorrelated signal provided by the control parameters. It should be noted that the gain factors used in the coding means 103 to 106 are not limited to a positive value. They can also be a negative value.

15

[0047] The output of means 107 and 108 for adding sub-band signals is introduced to means 109 and 110 to provide a signal in the time domain. The 109 output corresponds to the left channel of the reconstructed stereo signal, and the 110 output corresponds to the right channel of the reconstructed stereo signal. In the embodiment described in this case, the same decorator is used for both output channels, while the

twenty Means for adding the signal related to the unprocessed signal are different for the two output channels. The presently described embodiment therefore ensures that the two output signals can be identical as well as being completely decorrelated, depending on the control data provided to the means for adjusting the signal levels, and the control data provided to the means to add the signals.

25

[0048] A block diagram of the means for providing a correlated signal is shown in Figure 2. The input subband signal is introduced to the means 201 to filter a subband signal. In the embodiment of the present invention currently described, the filtering step is a reverberation unit that incorporates a total pass filtering. The filter coefficients used are provided

30 by means 202 to provide filter coefficients. The subband index of the currently processed subband signal is introduced at 202. In one embodiment of the present invention, different filtering coefficients are calculated based on the subband index provided at 202. The filtering stage at 201, it depends on delayed samples of the input subband signal as well as delayed samples of intermediate signals in the filtering process.

35

[0049] An essential feature of the present invention is that they are provided by means of 203 means to provide a whole sub-band sample delay and a fractional sub-band sample delay. The output of 201 is introduced in means to regulate the level of the sub-band signal 204, and also in means 205 to estimate signal characteristics of the sub-band signal. In a preferred embodiment of the present

40 invention, the estimated characteristic is the transient behavior of the sub-band signal. In this embodiment, a detected transient is signaled to the means 204 for regulating the level of the sub-band signal, so that the level of the signal is reduced during the transient steps. The output of 204 is the decorrelated signal introduced in 104 and 105 of Figure 1.

Four. Five [0050] Figure 3 shows the bank of individual analysis filters and the two filter banks of

synthesis. The analysis filter bank 301 operates on the mono input signal, while the synthesis filter banks 302 and 303 operate on the reconstructed stereo signals.

[0051] Therefore, Figure 1 shows the inventive apparatus for generating a decorrelation signal that is

fifty indicated by reference 102. As shown in Figures 1 or 3, this apparatus includes means for providing a series of sub-band signals, where a sub-band signal includes the sequence of at least two sub-samples. band, the sequence of sub-band samples representing a bandwidth of the sub-band signal that is less than a bandwidth of the input signal. Each sub-band signal is introduced into the filtering means 201. Each filter means 201 includes a reverberation filter, so that a series of reverberated sub-band signals is obtained, wherein said series of reverberated sub-band signals together represent the decorrelation signal. Preferably, as shown in Figure 2, there may be postprocessing at the subband level of reverberated subband signals, which is carried out by block 204, which is controlled by block 205.

[0052] Each reverberation filter is configured with a certain delay, and preferably a fractional delay, and each reverberation filter has several filter coefficients which, as shown in Figure 2, depend on the sub-band index. This means that it is preferable to use the same delay for each subband, but to use different sets of filter coefficients for the different subbands. This is symbolized by

5 means 203 and 202 of Figure 2, although in this case it should be mentioned that delays and filter coefficients are preferably determined in a fixed manner when a decorrelation device is shipped, where delays and filter coefficients can be determined empirically using hearing tests, etc.

[0053] A multi-channel decoder is shown in Figure 1, and includes the inventive apparatus for

10 generate the correlation signal, which is indicated as 102 in Figure 1. The multichannel decoder shown in the

Figure 1 is for decoding a mono signal and an associated coherence measurement between channels, the coherence measurement between channels representing a coherence between a series of original channels, where the mono signal is obtained from said series of original channels. Block 102 of Figure 1 constitutes a generator for generating a decorrelation signal for the mono signal. Blocks 103, 104, 105, 106, 107 and 108 constitute a

fifteen mixer to mix the mono signal and the decorrelation signal, in accordance with the first mixing mode in order to obtain a first decoded output signal, and in accordance with the second mixing module in order to obtain a second decoded output signal , where the mixer is operative to determine the first mixing module and the second mixing module based on the measurement of coherence between channels transmitted as lateral information to the mono signal.

20

[0054] Preferably, the mixer is operative for mixing in a sub-band domain, based on consistency measurements between independent channels for different sub-bands. In this case, the multi-channel decoder further comprises means 109 and 110 for transforming the first and second decoded output signals, from the sub-band domain to the time domain, in order to obtain a first output signal.

25 decoded and a second decoded output signal in the time domain. Therefore, the inventive means 102 for generating a decorrelation signal and the inventive multichannel decoder shown in Figure 1, operate in the sub-band domain and carry out, as a last step, a sub domain transformation -bands to time domain.

30 [0055] Depending on the actual situation, the inventive device can be implemented in physical equipment or

in software, or in an unalterable software that includes hardware components and software components. When partially or fully implemented in software, the invention is also a computer program with computer-readable code for carrying out the inventive procedures when running on a computer.

35