Advanced processing based on a complex-exponential-modulated filterbank

Description

TECHNICAL FIELD [0001] The present invention relates to audio source coding systems, but the same methods can also be used in many other technical fields. Various techniques are presented that are useful in audio coding systems using parametric representations of stereo properties.

BACKGROUND OF THE INVENTION AND BACKGROUND ART [0002] The present invention relates to parametric coding of an image of a stereo audio signal. Typical parameters used to describe the properties of a stereo image are inter-channel intensity difference (IID), inter-channel time difference (ITD) and inter-channel coherence (IC) -channel coherence). In order to reconstruct the stereo image based on these parameters, a method is needed that allows reconstruction of the correct level of correlation between the two channels according to the IC parameter. This is achieved by means of a de-correlation method.

[0003] There are several methods available for creating de-correlated signals. In an ideal situation, a linear time invariant (LTI) function with all-band frequency response is desirable. One obvious way to achieve this is to use a fixed delay. However, the use of delay or any other LTI all-pass functions will trigger a non-all-response response when the raw signal is added. In the event of a delay, the result will be a typical comb filter. The comb filter often produces an unwanted "metallic" sound that, even if the stereo widening effect is efficient, greatly reduces the naturalness of the original.

[0004] Methods in the frequency domain of generating a de-correlated signal by adding a random sequence to the IID value along the frequency axis in which different sequences for different audio channels are used are also known in the art. One of the problems with decorrelation in the frequency domain by modifying the random sequence is the introduction of leading echoes. Subjective tests have shown that for non-stationary signals, leading echoes are much more annoying than ordinary echoes, which is well justified by established psychoacoustic principles. This problem could be reduced by dynamically adapting the transform size to the signal characteristics in terms of transient content. However, changing the transform size is always a difficult (i.e. binary) decision that affects the entire signal bandwidth and which can be difficult to perform reliably.

[0005] In US Patent Application US 2003/0219130 A1, audio coding based on coherence and synthesis is disclosed. In particular, the auditory stage is synthesized from a monophonic audio signal by modifying, for each critical band, an auditory stage parameter such as interaural level difference (ILD) and / or intervertebral time difference (ITD) (ITD). inter-aural time difference) for each subband in the critical band, where the modification is based on the average estimated coherence for the critical band. Modification based on coherence creates auditory scenes with an object width that more accurately matches the width of the objects in the original auditory input stage. Stereo parameters are well-known BCC parameters, where BCC stands for binaural cue coding. When generating two different de-correlated output channels, the frequency coefficients obtained by discrete Fourier transform are grouped together in one critical band. Based on the measure of inter-channel coherence, the weighting coefficients are multiplied by the sequence of random pseu, which is preferably chosen in such a way that the variance is approximately constant for all critical bands, and in each critical band the average is equal to "0. The same sequence is used for the spectral coefficients of each other frame.

[0006] US Patent Application US 6005946 discloses a number of different types of signals (at least two) that are created from a monophonic input signal by all-pass filtering per total delay subband.

SUMMARY OF THE INVENTION [0007] The object of the present invention is to provide a decoding concept for parametric coded multi-channel signals or a coding concept for generating such signals, which leads to high audio quality and high coding efficiency.

[0008] This object is achieved by means of a device for generating a de-correlation signal according to claim 1, a method of generating a de-correlation signal according to claim 7, or a computer program according to claim 9.

[0009] The present invention is based on the finding that a good de-correlation signal is obtained on the decoding side for generating the first and second multi-channel signal channels based on a monophonic input signal when a reverberation filter is used which introduces total or preferably fractional delay to the input signal . Importantly, this reverberation filter is not applied to the entire input signal. Instead, several reverberation filters are used for several subbands of the original input signal, i.e. a monaural signal, so that reverberation filtering using reverberation filters is not applied in the time domain or in the frequency domain, i.e. in the field obtained by using transformation Fourier. According to the invention, the reverberation filtering using the reverberation filters for subbands is carried out individually in the subband domain.

[0010] The subband signal comprises a sequence of at least two subband samples, the sequence of subband samples representing the bandwidth of the subband signal which is smaller than the bandwidth of the input signal. Naturally, the bandwidth of the subband signal is greater than the bandwidth assigned to the frequency factor obtained by Fourier transform. The subband signals are preferably generated by means of a filter bank having, for example, 32 or 64 filter bank channels, while the FFT transformation for the same example will contain 1024 or 2048 frequency coefficients, i.e. frequency channels.

[0011] The subband signals are subband signals obtained by subband filtering of the input signal sample block. Because reverberation filtering is not applied to the entire signal but is used subband, the "metallic sound caused by comb filtering is avoided.

[0012] In cases where the sample period between two consecutive subband samples is too large to provide a good listening experience on the decoder side, fractional delays are used in the reverberation filter, for example a delay between 0.1 and 0.9, and preferably from 0.2 to 0.8 sampling period of the subband signal. It should be noted that for critical sampling and when 64 subband signals are generated using a filter bank having 64 filter bank channels, the sampling period in the subband signal is 64 times the sampling period of the original input signal.

[0013] It should be noted that in this case delays are an integral part of the filtering process used in the reverberation device. The output signal consists of many delayed versions of the input signal. It is preferable to delay the signals by fractions of the subband sampling period to obtain a good reverberation device in the subband domain.

[0014] 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. Regardless, the filter coefficients are different for each subband. The use of IIR filters is beneficial. Depending on the actual situation, the fractional delay and filter coefficients for different filters can be determined experimentally using listening tests.

[0015] The subbands filtered by means of a set of reverberation filters create a de-correlation signal to be mixed with the original input signal, i.e. a mono signal to obtain a decoded left channel and a decoded right channel. This mixing of the decorrelation signal with the original signal is implemented based on the inter-channel coherence parameter sent together with the parametrically coded signal. In order to obtain different left and right channels, i.e. different first and second channels, mixing the de-correlation signal with the mono signal to obtain the first output channel is different from mixing the de-correlated signal with the mono signal to obtain the second output channel.

[0016] In order to obtain greater efficiency on the coding side, multi-channel coding is carried out using adaptive stereo parameter set. To this end, the encoder includes, in addition to the means for calculating the mono signal and in addition to the means for generating the set of stereo parameters, means for determining the validity of the sets of stereo parameters for the successive parts of the left and right channels. Preferably, the determining means operate to activate the generating means when it is determined that the set of stereo parameters is no longer valid, so that the second set of stereo parameters is calculated for a portion of the left and right channels starting from the second time limit. The second time limit is also set by means of determining validity.

[0017] Thus, the encoded output signal comprises a monophonic signal, a first set of stereo parameters and a first time limit associated with the first set of parameters and a second set of stereo parameters and a second time limit associated with the second set of stereo parameters. On the decoding side, the decoder will use a valid set of stereo parameters until it reaches a new time limit. After reaching a new time limit, decoding operations are performed using a new set of stereo parameters.

[0018] Compared with prior art methods that utilize block processing, and thus, blocking of stereo parameter sets, adaptive determination of stereo parameter sets of the invention for different encoder-defined time boundaries, on the one hand, provides high coding efficiency and high quality coding on the other hand. This is because for relatively stationary signals, the same set of stereo parameters can be used for multiple blocks of mono signal samples without introducing audible errors. On the other hand, when considering non-stationary signals, the adaptive stereo parameter setting according to the invention provides better time resolution, so that each part of the signal has its optimal set of stereo parameters.

[0019] The present invention provides a solution to the problems of the prior art by using the reverberation unit as a decorrelator implemented with fractional delay lines in the filter bank and using adaptive level of the de-correlated reverberation signal.

[0020] Next, several aspects that are not part of the present invention will be briefly outlined.

[0021] One aspect of the invention is a method of delaying a signal by: filtering a time domain real value signal through part of the analysis of a complex filter bank; modifying subband signals with complex values ​​obtained from filtering; and filtering the complex subband signals with complex values ​​by a portion of the filter bank synthesis; and taking the real portion of the complex-time output signal, where the output signal is the sum of the signals obtained from the synthesis filtering.

[0022] A hymn aspect of the invention is a method of modifying subband signals with complex values ​​by filtering each subband signal with complex values ​​by means of a filter with a finite impulse response with complex values, where a filter with a finite impulse response for the number n of the subband is defined as Fourier transform for signals discrete in form. (exp (—ίπ (η + 1/2) τ) 6_τ(ω), for even n;

(εχρ (-ίπ (η + 1/2) τ) 6_τ(ω + π), for n odd., _gc]_fromj_e parameter T = T / L and where the synthesis filter bank has L subbands and the desired delay T is measured in units of the output signal sample.

[0023] A hymn aspect of the invention is a method of modifying complex subband signals by filtering, where G_T(what) approximately meets ν_τ(Ω) θ_τ(ω) + ν_τ(Ω + π) Ο_τ(ω + π) = \, where V_T(co) is a Fourier transformation for signals ν_τ(k) = A i^k^ p (l} p (lT -Lk \ discrete sequences / ip (l) is a prototype filter of said complex filter bank, while A is a suitable factor for real normalization.

[0024] This aspect of the invention is a method of modifying complex subband signals by filtering, where G_T(a>) meets Ο_τ(-Ω) = Ο_τ(ω + π) ^ ', so that the even-indexed impulse response samples are only real values, and the odd-indexed impulse response samples are only imaginary values.

[0025] Another aspect of the invention is a method of coding the stereo properties of an input signal in a encoder, wherein the calculated time grid parameters describe the position in time for each set of stereo parameters, where the number of sets of stereo parameters is random, and in the decoder, the synthesis of the stereo parameter according to this time grid.

[0026] Another aspect of the invention is a method of coding the stereo properties of an input signal, where the time location for the first set of stereo parameters is, in the case where the time signal for the set of stereo parameters coincides with the beginning of the frame, clearly signaled instead of signaling the time indicator.

[0027] This aspect of the invention is a method of generating stereo de-correlation for parametric stereo reconstruction in a decoder using an artificial reverberation process to synthesize an additional signal.

[0028] A hymn aspect of the invention is a method of generating stereo de-correlation for parametric stereo reconstruction in a decoder where the reverberation process is performed in an exponentially modulated filter bank using phase delay control on each filter bank channel.

[0029] A hymn aspect of the invention is a method of generating stereo de-correlation for parametric stereo reconstruction in a decoder, wherein the reverberation process uses a detector designed to find signals, where the end part of the reverberation may be undesirable and allows the suppression part of the reverberation to be suppressed or removed.

BRIEF DESCRIPTION OF THE DRAWINGS [0030] The present invention will now be described in the form of illustrative examples, without limiting the scope of the invention, with reference to the accompanying drawings, in which:

Fig. 1 shows a block diagram of an apparatus according to the invention;

Fig. 2 shows a block diagram of a means for generating a de-correlated signal;

Fig. 3 shows a single channel analysis and synthesis of a pair of stereo channels based on the reconstructed stereo subband signals of the present invention;

Fig. 4 is a block diagram of the division of parametric sets of stereo parameters into time segments based on signal properties; and

Fig. 5 shows an example of the division of parametric sets of stereo parameters into time segments based on signal properties.

DESCRIPTION OF THE PREFERRED EMBODIMENTS [0031] The embodiments described below only illustrate the principles of the present invention for parametric stereo coding. It is understood that modifications and changes to the systems and details described herein are obvious to those skilled in the art. It is therefore intended that the invention be limited only by the scope of the following claims, and not by the particular details presented by way of description and explanation of the present embodiments.

[0032] A signal delay by a fraction of a sample can be achieved using several interpolation methods known in the art. However, there are special cases when the original signal is available as oversamplified samples with complex values. The implementation of the fractional delay in the bank qmf only by applying a phase delay by a factor, for each channel qmf corresponding to a constant time delay, brings severe artifacts.

[0033] This can be efficiently avoided by using a compensating filter according to an innovative approach enabling high quality approximation of random delays in any complex-exponential-modulated filter bank. A detailed description is provided below.

Continuous time model [0034] To facilitate calculations, the L-band exponentially modulated complex values ​​filter bank will be mapped here by the continuous-windowed transformation using waveform synthesis

<img file="PL3244638T3_D0001.tif" />

where n, k are integers at n&gt; 0 and θ is the phase constant. The results for discrete time signals are obtained by appropriate sampling of the variable t at 1 / L intervals. It is assumed that the v (t) window with real values ​​is selected, such that for x (t) signals with real values ​​with high accuracy the relationship is true

<img file="PL3244638T3_D0002.tif" />

if

<img file="PL3244638T3_D0003.tif" />

where * is the complex join. It is also assumed that v (t) is essentially band limited to the frequency range [-π, π]. Considering the modification of each n-frequency band by filtering discrete time Cn (k) samples with a hn (k) impulse response filter,

1. ») ^^ · (4) l

[0035] Then modified synthesis = 2Re £_n (to_nk(t)&gt;

^ «= 0 λ = -00 _&gt; can be calculated in the frequency domain as y (co) = Η (ω) χ (ω), where βω) means Fourier transforms f (t) and

Η (ω) = Σ //,(c.9)|ćUo-tG? + I / 2))² .

(5) (7) (6) w = -co [0036] In this case, ^^η(^ω) ^_ Ek% (/ c) exp (-ikćo) j_es^ Fourier transform for discrete signals of the filter applied in the n frequency band for n&gt; 0 i

Η_η(ω) = (-ω) * for n &lt;0. (8) [0037] It should be noted here that the special case H_n(cd) = l leads to H (co) = lw (7) due to the special window construction v (t). hmy interesting case here is H_in(m) = exp (-zm), which gives H (m) = exp (-zm), so that y (f) = x (tl).

Proposed solution [0038] In order to obtain a delay of size r such that y (t) = x (tT), the problem is to design filters Η_η(ω) for n&gt; 0, so that

Η (ω) = exp (-zno), (9) where Η (ω) is defined by (7) and (8). A special solution proposed here is the use of iexp filters (—ίπ (η + 1/2) τ) filtr_τ(ω), for even n;

(exp (—ίπ (η + 1/2) τ) ή_τ(ω + π), for n odd.

[0039] Here G_T(-M) = G_T(m + 7r) * implies consistency with (8) for all n. Introduction (10) on the right side of equation (7) gives

Η (ω) - οχρ (- / ωτ) [X (ω) 6) (ω) + F_T(ω + π) θ_τ(ω + π)] (11) where

Κ (^ω) = “^π(^{2λ7 + 1/2})) z ó (m) = exp (zrm) | v (m) |². Basic calculations show that F ^ m) is a Fourier transform for discrete signals v_T(k) = i^k jv (t) v (t-τ -k) dt. (12) [0040] Very good approximations of an ideal delay can be obtained by solving a linear system

F / ®) (j_r(®) + F_r(® + π) 6_τ(ω + π) = 1 (13) by the least squares method using the FIR filter

G, (®) = ​​Fil, vs, W exp (-a-ffl).

In the range of filter coefficients, equation (13) can be written

2Fv_T(24 - /) g_T(/) = 5 [ą, (14) l

where ó [ł] = l for k = Q and ó [ł] = 0 for k / 0 [0041] In the case of a discrete-time F-band filter bank with a prototype filter p (k), the obtained delay in sample units is L_T and the calculation (12) is replaced by

Κ ^) = ί * Σρ (1) ρ (1-Τ-Ιά), (15) and

where T is the integer nearest to L_T. Here p (k) is extended by zeros outside the operating range. For a prototype filter with a finite length, only a finite number v_T(k) is different from zero, and (14) is a system of linear equations. Number of unknowns g_T(k) is usually chosen as a small number. In the design of good QMF filter banks, 3-4 members already provide very good delay performance. In addition, the dependence of terms g_T(k) the filter from the delay parameter τ can often be effectively modeled by low-degree polynomials.

Adaptive time grid signaling for stereo parameters [0042] Parametric stereo systems always lead to trade-offs in terms of limited time or frequency resolution to minimize the amount of data transmitted. However, it is known from psychoacoustics that some spatial signals may be more important than others, which leads to the possibility of removing less important signals. Therefore, the time resolution does not have to be constant. A large gain in throughput can be obtained by enabling synchronization of the time grid with spatial signals. This can easily be achieved by sending a variable number of parameter sets for each data frame that corresponds to a fixed-size time segment. In order to synchronize parameter sets with the corresponding spatial signals, additional time grid data describing the position in time for each parameter set must be sent. The resolution of these time indicators can be chosen as quite low in order to keep the total amount of data minimized. A special case in which the time signal for the parameter set coincides with the beginning of the frame could be signaled clearly to avoid sending this time indicator.

[0043] Fig. 4 shows an apparatus according to the invention for performing parameter analysis for time segments having variable and signal-dependent time limits. The device according to the invention comprises means 401 for dividing the input signal into one or several time segments. Time limits that separate time segments are provided by means 402. Means 402 use a detector specifically designed to acquire spatial signals that are important for deciding where to set time limits. Means 401 send the entire input signal divided into one or several time segments. This output is input to means 403 for separate parameter analysis for each time segment. Means 403 send one set of parameters to the analyzed time segment.

[0044] Fig. 5 shows an example of how a time grid generator may behave in the case of a hypothetical input signal. This example uses one set of parameters per data frame if there is no other information about the time limits. Hence, when no other time limit information is present, inherent time frames of the data frame are used. The time limits shown in Fig. 5 are derived from the means 402 in Fig. 4. The time segments shown in Fig. 5 are provided by means 401 in Fig. 4.

[0045] The apparatus for encoding the stereo signal to obtain the monophonic output signal and the set of stereo parameters comprises means for calculating the monophonic signal by combining the left and right channels of these stereo signals in weighted summation. In addition, means 403 generate a first set of stereo parameters using the left channel part and the right channel part, the parts beginning at the first time limit being connected to the means for determining the validity of the first set of stereo parameters for the subsequent left and right channel parts.

[0046] The fixing means are formed together by means 402 and 401 in Fig. 1.

[0047] In particular, the determining means are capable of generating a second time limit and activating the generating means when it is determined that the first set of stereo parameters is no longer valid, such that a second set of stereo parameters is generated for a portion of the left and right channels starting from the second time limit.

[0048] Not shown in Fig. 4 are the means for outputting the mono signal, the first set of stereo parameters 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 as the parametrically coded stereo signal. The means for determining the validity of a set of stereo parameters may include a transient detector, since there is a high probability that a new stereo parameter must be generated after the transient, since the signal has changed significantly in shape. Alternatively, validation means may include a synthesis analysis device that is adapted to decode a mono signal and a set of stereo parameters to obtain a decoded left and decoded right channel in order to compare a decoded left channel and a decoded right channel with a left channel and a right channel and to activate the means to generate, when the decoded left channel and the decoded right channel differ from the left channel and right channel by more than the set threshold.

[0049] Data frame 1: The time segment corresponding to the set of 1 parameters begins at the beginning of the data frame 1 because there is no other time limit information in this data frame.

[0050] Data frame 2: Two time limits occur in this data frame. The time segment corresponding to the set of 2 parameters begins at the first time limit in this data frame. The time segment corresponding to the set of 3 parameters begins at the second time limit in this data frame.

[0051] Data frame 3: One time border occurs in this data frame. The time segment corresponding to the set of 4 parameters begins at the time limit in this data frame.

[0052] Data frame 4: One time border occurs in this data frame. This time limit coincides with the initial limit of the data frame 4 and does not need to be signaled because it is supported by default. Hence, the time limit signal can be removed. The time segment corresponding to the set of 5 parameters begins at the beginning of the data frame 4, even without signaling this time limit.

The use of artificial reverberation as a method of de-correlation for parametric stereo reconstruction [0053] One necessary stage in the implementation of stereo synthesis in a parametric stereo system is to reduce the coherence between the left and right channels to create a stereo image width. This can be done by adding a filtered version of the original monophonic signal to the additional signal, where the additional signal and monophonic signal are defined by monophonic = (left + right) / 2, respectively, and

additional = (left - right) / 2.

[0054] In order not to change the color too much, the filter should preferably be an all-pass filter. One effective approach is to use similar all-pass filters used in artificial reverberation processes. Artificial reverberation algorithms usually require high time resolution to provide an impulse response that is sufficiently diffused in time. The artificial reverberation algorithm based on a complex filter bank, such as a complex qmf bank, has great advantages. The filter bank provides excellent opportunities for reverberation properties to be selected selectively for frequencies in the field of, for example, correction of reverberation, decay time, density and color. However, filter bank implementations usually change the time resolution for higher frequency resolutions, which usually makes it difficult to implement a reverberation process with a sufficiently smooth time course. To deal with this problem, the new method will be to use approximation with fractional delay by applying only phase delay by factor, for each channel qmf corresponding to a fixed time delay. This simple method of fractional delay introduces a serious blur of time, which fortunately in this case is very desirable. Time blur contributes to time diffusion, which is highly desirable for reverberation algorithms, and becomes larger as the phase delay approaches pi / 2 or -pi / 2.

[0055] Artificial reverberation processes are naturally processes with an infinite impulse response and offer natural exponential decay. [PCT / SE02 / 01372] indicates that if a reverberation unit is used to generate a stereo signal, reverberation loss can sometimes be undesirable as soon as the sound ends. However, these unwanted reverberation end portions can easily be suppressed or completely removed only by changing the reverberation signal gain. A detector designed to find the end parts of the sound can be used for this purpose. If the reverberation unit generates artifacts in certain specific signals, e.g. transients, the detector for these signals can also be used to suppress them.

[0056] Fig. 1 shows a device according to the invention for a method of de-correlation of signals used in a parametric stereo system. The device according to the invention comprises means 101 for providing a plurality of subband signals. The delivery means can be a complex QMF filter bank in which each signal is associated with a subband index.

[0057] The subband signals output to the means 101 of Fig. 1 are input to the input of the means 102 to provide a de-correlated signal 102 and to the input of the means 103 and 106 to modify the subband signal. The output of 102 is fed to the input of means 104 and 105 to modify the signal, and the output of 103, 104, 105 and 106 is input to the input of means 107 and 108 to add subband signals.

[0058] In the embodiment of the invention described herein, means 103, 104, 105 and 106 for modifying the subband signals match the level of the de-correlated signal and the raw signal, which is output 101 by multiplying the subband signal by a gain factor, so that each sum of each pair gives a signal with the size of the de-correlated signal set by the control parameters. It should be noted that the gain factors used in measures 103-106 to modify are not limited to a positive value. It can also be a negative value.

[0059] The output of the means for adding subband signals 107 and 108 is fed to the input of the means to provide a time domain signal 109 and 110. The output from 109 corresponds to the left channel of the reconstructed stereo signal, and the output from 110 corresponds to the right channel of the reconstructed stereo signal. In the embodiment described here, the same decorrelator is used for both output channels, and the means for adding the de-correlated signal to the raw signal are separate for the two output channels. The embodiment described here ensures, therefore, that both output signals can be identical and at the same time completely de-correlated, depending on the control data supplied to the means for adjusting signal levels and the control data provided to the means for adding signals.

[0060] Fig. 2 is a block diagram of means for providing a de-correlated signal. The input subband signal is fed to the input of means 201 to filter the subband signal. In the embodiment of the present invention described herein, the filtering step is a reverberation unit comprising all-pass filtering. The filter coefficients used are provided by means 202 for providing filter coefficients. The subband index of the currently processed subband signal is fed to the input to 202. In one embodiment of the present invention, various filter coefficients are calculated based on the subband index provided to 202. The filtering step in 201 is based on delayed samples of the input subband signal as well as on delayed samples of intermediate signals in the filtering procedure.

[0061] An important feature of the present invention is that the means for providing the total delay of the subband sample and the fractional delay of the subband sample are provided by 203. The output of 201 is applied to the input of the means 204 to adjust the level of the subband signal as well as to the input of the means 205 in to estimate subband signal properties. In a preferred embodiment of the present invention, the estimated properties are transient subband signal behavior. In this embodiment, the detected transient is signaled to the means 204 to adjust the level of the subband signal, such that the signal level is reduced during transient transitions. The output of 204 is the de-correlated input for 104 and 105 in Fig. 1.

[0062] Fig. 3 shows a single analysis filter bank and two synthesis filter banks. 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.

[0063] Fig. 1 therefore shows a device according to the invention for generating a de-correlation signal, which is indicated by 102. As shown in Figs. 1 or 3, this device comprises means for providing a plurality of subband signals, wherein the subband signal comprises a sequence of at least of two subband samples, the sequence of subband samples represents the bandwidth of the subband signal that is smaller than the bandwidth of the input signal. Each subband signal is input to the filter means 201. Each filter means 201 includes a reverberation filter, such that a plurality of subband reverberation signals are obtained, with the plurality of subband reverberation signals together representing the de-correlated signal. Preferably, as shown in fig. 2, subband post-band reverberation processing may take place, which is performed by block 204, which is controlled by block 205.

[0064] Each reverberation filter is set to a certain delay, preferably a fractional delay, as well as each reverberation filter has several filter coefficients, which, as shown in Figure 2, depend on the subband index. This means that it is preferable to use the same delay for each subband, but different sets of filter coefficients for different subbands. This is symbolized by means 203 and 202 in Fig. 2, although it should be mentioned that delays and filter coefficients are preferably fixed when sending a de-correlation device, delays and filter coefficients can be determined empirically using listening tests and the like.

[0065] In Fig. 1, a multi-channel decoder is shown and includes a device according to the invention for generating a correlation signal, which is denoted by 102 in Fig. 1. The multi-channel decoder shown in Fig. 1 is used for decoding a monophonic signal and an associated measure of inter-channel coherence, where the measure of inter-channel coherence represents the coherence between many original channels, with the monophonic signal being obtained from many original channels. Block 102 in Fig. 1 is a generator for generating the de-correlation signal for a mono signal. Blocks 103, 104, 105, 106 and 107 and 108 constitute a mixer for mixing the monophonic signal and the de-correlation signal according to the first mixing mode to obtain the first decoded output signal and according to the second mixing mode to obtain the second decoded output signal, wherein the mixer is capable of determining a first mixing mode and a second mixing mode based on a measure of inter-channel coherence sent as additional information to a mono signal.

[0066] The mixer is preferably capable of mixing in the subband domain based on separate inter-channel coherence measures for different subbands. In this case, the multi-channel decoder further includes means 109 and 110 for converting the first and second decoded subband output signal into the time domain to obtain the first decoded output signal and the second decoded output signal in the time domain. Thus, means 102 for generating the de-correlation signal according to the invention and the multi-channel decoder according to the invention, shown in Fig. 1, operate in the subband domain and perform, in the last step, the conversion from the subband domain to the time domain.

[0067] Depending on the actual situation, the device according to the invention may be implemented in hardware or in software or as firmware including firmware components and software components. When implementation in the form of a program partially or in full, the invention is also a computer program having a computer readable code for implementing the methods of the invention when the program is running on a computer.