RU2834349C2 - Audio decorrelator, processing system and method for decorrelation of audio signal - Google Patents

Abstract

FIELD: processing data.

SUBSTANCE: present invention relates to an audio signal decorrelator. Decorrelator comprises a plurality of delay units, wherein each delay unit is configured to receive a portion of the frequency representation based on an audio signal, wherein each delay unit is configured to delay the received portion to provide a delayed portion. Decorrelator comprises an envelope generator configured to receive and combine signals based on delayed portions of the frequency representation. Envelope generator receives a frequency representation of the audio signal and is configured to adjust the energy of the delayed portions with respect to the frequency representation of the audio signal. Envelope generator is configured to provide a frequency representation of the combined shape. Transient parts of the signal are handled by the adapted operation of the decorrelator.

EFFECT: enabling low processing delay and/or decorrelation with low complexity and high perceptual quality, particularly when processing signals containing transients.

41 cl, 10 dwg

Description

The present invention relates to a decorrelator for an audio signal, to a processing system having such a decorrelator, to a decorrelation method and to a computer program product. The present invention relates in particular to an audio signal decorrelator.

In perceptual audio coding, decorrelators are an important building block for parametric spatial audio coding. Known solutions are decorrelators known from parametric spatial audio coding, such as the parametric stereo approach or the MPEG surround standard. Decorrelators, as described in [1] or in [2], use computationally expensive time-domain reverb filters with long impulse response. Decorrelators, such as those described in [3] or [4], require the use of a quadrature mirror filter bank (QMF) with significant processing delay and computationally expensive lattice filters.

There is therefore a need for a decorrelator, a processing system having such a decorrelator, and a method for decorrelating portions of an audio signal that enables decorrelation with low processing latency and/or low computational complexity.

The object of the present invention is to provide a decorrelator, a processing system and a method for decorrelation that enable low processing latency and/or decorrelation with low complexity and high perceptual quality, in particular when processing signals containing transients.

This objective is achieved by means of the subject matter of the invention defined in the independent claims of the invention.

The result of the present invention is that dividing the frequency representation into multiple parts and processing, i.e. delaying each of the parts by a separate delay module, enables low processing delay, since the calculation of different parts can be performed in parallel. At the same time, such operations in the frequency domain require low computational complexity.

According to an embodiment, the decorrelator comprises a plurality of delay modules, wherein each delay module is configured to receive a part of a frequency representation based on an audio signal, wherein each delay module is configured to delay the received part in order to provide a delayed part. The decorrelator comprises an envelope shaper configured to receive and combine signals based on delayed parts of the frequency representation, with the ability to receive a frequency representation of the audio signal, with the ability to regulate the energy of the delayed parts in relation to the frequency representation of the audio signal and with the ability to provide a frequency representation of a combined form.

According to an embodiment, the different parts of the frequency representation contain the same or different number of frequency bins. In this case, the same number of frequency bins can provide the possibility of the same processing time, and a different number of frequency bins can provide the possibility of adaptation to the requirements of the application.

According to an embodiment, the decorrelator comprises a phase shifter, configured to phase shift the frequency representation of the audio signal or to phase shift the audio signal in the time domain to obtain a phase-shifted audio signal. The phase shift can provide the possibility of perceptible reverberation and, due to this, high sound quality.

According to an embodiment, the phase shifter is configured to phase shift the frequency representation of the audio signal and comprises a plurality of all-pass filters, wherein each all-pass filter is configured to phase shift an associated part of the frequency representation of the audio signal. Thus, the all-pass filter can be associated and adapted to the corresponding part of the audio signal, which makes it possible to provide high overall sound quality.

According to an embodiment, the all-pass filter of the plurality of all-pass filters comprises a set of all-pass filter structures connected in series with each other, i.e., IIR Schroeder filters are used. The all-pass filter structures are designed to provide different time delays. Alternatively or additionally, the all-pass filter structures comprise a nested all-pass filter structure.

According to an embodiment, the number of all-pass filter structures and/or the structure scheme of the all-pass filters are equivalent or different between different all-pass filters. This allows for high flexibility of the decorrelator.

According to an embodiment, the various time delays are based on a prime multiple of the local sampling frequency used to obtain the frequency representation of the audio signal. This enables high perceived sound quality.

According to an embodiment, the set of all-pass filter structures comprises four all-pass filter structures and they are configured to provide a delay of 1, 2, 3 and 5 time units. Such a time unit may be based on the size of the block of the transformation to the frequency domain. For example, when using a block size of 256 with 50% overlap, a time unit may result in 128 samples @ 48 kHz = 2.7 ms. Reasonable other time units, for example, can be 32 or 64 samples or other values. The time units are preferably short enough to allow sufficient time resolution in the subsequent time/frequency envelope shaping. In an alternative solution, a delay of 1, 3, 5 and 7 is provided by four all-pass filter structures. This allows for no overlaps in the time domain.

According to an embodiment, the gain of the all-pass filter is adapted to a value with an absolute value, i.e. with positive or negative values, of 0.7 within a tolerance range. The tolerance range is, for example, 20%, 10% or 5%.

According to an embodiment, the phase shifter is configured to phase shift the audio signal in the time domain, wherein the phase shifter comprises a set of all-pass filter structures connected in series with each other, wherein the all-pass filter structures are configured to provide different time delays. Alternatively or additionally, the all-pass filter structures comprise a nested all-pass filter structure.

According to an embodiment, the various all-pass time delays are based on a prime number that is a multiple of the reciprocal of the sampling frequency used to obtain the frequency representation of the audio signal. Similar to the frequency domain, a corresponding advantage can also be obtained in the time domain. In the time domain, the various time delays can be based on a prime number that is obtained by multiplying each of a set of minimal prime numbers, such as 1, 2, 3 and 5 as one example set or 1, 3, 5 and 7 as another example set, by a downsampling factor used to form parts of the frequency representation of the audio signal, in order to obtain an intermediate result, and to use the next prime number relative to the intermediate result. The next prime number can be understood as the closest distance, such as to obtain the next higher or the next lower prime value. In this example, the values 131, 257, 383 and 641 may be obtained for the first set and 131, 383, 641 and 907 may be obtained for the second exemplary set. Here, one unit of time may be 1 sample. A sample may be associated with a sampling frequency of, for example, 48 kHz. In other embodiments, the sampling frequency may also be 44.1 kHz or 32 kHz or other values.

According to an embodiment, the decorrelator comprises a first conversion module for obtaining a frequency representation of the audio signal from the audio signal for the envelope shaper and comprises a second conversion module for obtaining a frequency representation from the reverberated audio signal, wherein parts of the frequency representation constitute parts of the frequency representation from the reverberated audio signal. This makes it possible to generate a usable signal formed directly in the decorrelator.

According to an embodiment, the decorrelator is configured to additionally implement the same and predetermined delay for a subset or all parts of the frequency representation. Thus, a delay that is equal for the corresponding parts or delay lines can also be typically applied in a common delay unit that enables simple delay modules in the corresponding delay lines for the associated part.

According to an embodiment, the delay modules associated with a spectral part of the plurality of delay modules are configured to delay the associated part of the frequency representation differently compared to the delay modules associated with other spectral parts. This enables high perceptual quality by processing different frequency parts differently.

According to an embodiment, the delay module is configured to delay parts of the frequency representation containing lower frequencies with a higher time delay compared to parts of the frequency representation containing higher frequencies.

According to an embodiment, the relationship between the different time delays is linear, logarithmic and/or based on rounding over subband samples. This enables high perceived quality.

According to an embodiment, the decorrelator comprises a conversion module for receiving and converting an audio signal or a reverberated version of the audio signal into parts by performing a time-block discrete Fourier transform (DFT) or a short-time Fourier transform (STFT), wherein the conversion module is configured to convert blocks having an overlap of 50% within a tolerance range. Such block-by-block conversion ensures the possibility of small delays of the corresponding received part and the possibility of parallel processing of different parts.

According to an embodiment, the envelope shaper is designed to operate in the subband region and with a time resolution of less than 4 milliseconds.

According to an embodiment, the decorrelator comprises a signal processing stage configured to receive a signal based on a frequency representation of a combined form, for example, as a mono signal and to process the mono signal into at least a stereo signal. This provides the possibility of improved perception by the listener.

According to an embodiment, the decorrelator comprises a signal processing stage configured to process a frequency representation of the combined form into at least a stereo signal and to model the extent of the source based on at least the stereo signal, for example in the frequency domain.

According to an embodiment, the processing system comprises a decorrelator as described herein and a processing stage for converting the mid/side decomposed signal into a left/right decomposed signal.

According to embodiments, the processing system may perform transient suppression in order to suppress echo signals, such as pre-echo and/or post-echo, caused by the transient. Such transient processing may comprise muting the output of the decorrelator and, accordingly, amplifying the output of the delay compensation module providing a part of the left/right decomposed signal and located in parallel with the decorrelator and connected to the processing stage.

According to an embodiment, the method comprises receiving a plurality of parts of a frequency representation based on an audio signal, delaying each of the received parts to provide a plurality of delayed parts, and receiving and combining signals based on the delayed parts of the frequency representation. The method comprises receiving a frequency representation of an audio signal and adjusting the energy of the delayed parts with respect to the frequency representation of the audio signal. The frequency representation of the combined form is provided.

According to an embodiment, a computer program or computer program product, or a non-transient data carrier having stored instructions to execute corresponding instructions, is provided for implementing this method, when executed on a computer.

Additional advantageous embodiments are given in dependent claims.

Preferred embodiments are described in more detail with reference to the accompanying drawings, in which:

Fig. 1 shows a block diagram of a decorrelator according to an embodiment;

Fig. 2 shows a block diagram of a decorrelator comprising a transform module for generating a frequency representation of a time domain signal, according to an embodiment;

Fig. 3 shows a block diagram of a decorrelation further comprising a pre-delay, according to an embodiment;

Fig. 4 shows a schematic block diagram of an all-pass filter according to an embodiment;

Fig. 5 shows a basic block diagram of a nested structure of all-pass filters according to an embodiment;

Fig. 6 shows a block diagram of a decorrelator comprising a phase shifter configured to operate in the time domain according to an embodiment;

Fig. 7 shows a block diagram of a decorrelator coupled to source extent modeling, according to an embodiment;

Fig. 8 shows a basic block diagram of a processing system according to an embodiment;

Fig. 9 shows a block diagram of a processing system configured to process a transient, according to an embodiment; and

Fig. 10 shows a basic block diagram of a method according to an embodiment.

Identical or equivalent elements, or elements with identical or equivalent functionality, are designated in the description below by identical or equivalent reference signs even when appearing in different drawings.

In the following description, many details are set forth to provide a more complete explanation of the embodiments of the present invention. However, it will be understood by those skilled in the art that the embodiments of the present invention can be practiced without these specific details. In other cases, known structures and devices are shown in block diagram form rather than in detail so as not to obscure the understanding of the embodiments of the present invention. In addition, the features of the various embodiments described below can be combined with each other unless otherwise expressly indicated.

Fig. 1 shows a basic block diagram of a decorrelator 10 according to an embodiment. The decorrelator 10 comprises at least two delay modules 12 ₁ -12 _n , where n>1. Although Fig. 1 illustrates two delay modules 12, this number is preferably higher, for example 4, 8, 16 or other values that are to be obtained by a power of 2, while the embodiments are not limited to such numbers. Thus, the embodiments may also comprise 3, 5, 7 or 9 delay modules 12. Each delay module is configured to receive an associated portion 14 ₁ -14 _n of a frequency representation 14 based on an audio signal. For example, the frequency representation 14 may be or may comprise a spectrum obtained by means of a Fourier transform, for example a discrete Fourier transform (DFT) or a short-time Fourier transform (STFT). The parts 14 ₁ -14 _n can be obtained, for example, as a frequency subband of the spectrum, i.e. a part of the representation in the frequency domain. For example, such a part 14 ₁ -14 _n can be obtained by using an appropriate window.

Each delay module 12 ₁ -12 _n is configured to delay the received portion 14 ₁ -14 _n in order to provide a delayed portion 14' ₁ -14' _n , i.e. to have a delay in the time domain.

The decorrelator 10 further comprises an envelope shaper 16 configured to receive signals based on the delayed portions 14' ₁ -14' _n . Such signals may be directly the delayed portions 14' ₁ -14' _n or processed versions thereof. The envelope shaper 16 is configured to combine the received signals. In addition, the envelope shaper is configured to receive a frequency representation 14 of the audio signal. The envelope shaper 16 is configured to regulate the energy of the delayed portions 14' ₁ -14' _n with respect to the frequency representation 14 of the audio signal. The envelope shaper 16 is configured to provide a frequency representation 18 of a combined form. In the frequency representation 18 of the combined form, the signals corresponding to the portion 14 ₁ -14 _n , resulting from them, respectively, can be decorrelated with respect to each other and/or with respect to the frequency representation 14.

Although the envelope shaper 16 is illustrated in such a way that it receives the combined frequency representation 14, as an alternative, the envelope shaper 16 may receive the corresponding information by receiving possibly non-delayed or normally processed portions 14 ₁ -14 _n .

Fig. 2 shows a basic block diagram of a decorrelator 20 according to an embodiment. The decorrelator 20 is configured to receive an audio signal 22. The decorrelator 20 may comprise a transform module 24 configured to generate a frequency representation 14 shown in Fig. 1. The transform module 24 may provide portions 14 ₁ -14 ₁₆ obtained by means of an exemplary STFT. For example, the frequency representation may comprise frequency bins in a total of 129. Alternatively, 128 bins may be used. For example, two types of digital Fourier transforms (DFT) may be used, so-called "even-stacked" and "odd-stacked". For example, as a "standard" DFT, an even-stacked version may be considered, having, in the given example, 129 frequency bands (127 complex, one real and one imaginary). The odd-stacked version may contain 128 (complex) frequency bands. Both transforms may be used in the embodiments described herein. The portions 14 ₁ - 14 ₁₆ may contain, partially or completely, the same or different numbers of bins. For example, the portion 14 ₁ may contain the first to the ninth bin, for example 9 bins. The portion 14 ₂ contains, for example, bins 10 - 19 and therefore bins in number 10. The adaptation or selection regarding the number of bins may be based on the sampling frequency, in the illustrated example 48 kHz, on the overlap, i.e., for example, 50%, and/or on the number of portions 14 ₁ - 14 ₁₆ that are to be generated. Portions 14 ₁ -14 ₁₆ may contain an equal or different number of frequency bins, so that some or all of portions 14 ₁ -14 ₁₆ may also be formed such that they contain the same number of frequency bins.

The decorrelator 20 further comprises a delay section 25 having delay lines 12 ₁ -12 ₁₆ , wherein each delay line 12 ₁ -12 ₁₆ is associated with one particular portion 14 ₁ -14 ₁₆ and is configured to receive said portion, its processed version, respectively. The delay modules 12 ₁ -12 ₁₆ can be associated with a corresponding spectral portion 14 ₁ -14 ₁₆ . Such a delay module 12 ₁ -12 ₁₆ can be configured to delay the associated portion of the frequency representation 14 differently compared to the delay modules associated with other spectral portions. Alternatively or additionally, the relationship between the different time delays can be one of linear, logarithmic and/or based on rounding over superband samples.

The decorrelator 20 further comprises a phase shifter 26 connected to the delay section 25, wherein the phase shifter 26 is configured to receive the delayed portions 14' ₁ -14' ₁₆ . The phase shift using the phase shifter 26 can provide the possibility of reverberation in the portions of the signal. However, according to embodiments, the sequence of the delay section 25 and the reverberation section 26 can also be changed in such a way that the corresponding portion 14 ₁ -14 ₁₆ can first be subjected to processing by the reverberation filter and delayed thereafter.

The phase shifter 26 can be configured to phase shift the frequency representation 14 of the audio signal, its processed, for example delayed, version. The phase shift can also be performed before converting the audio signal 22 into the frequency domain, the corresponding phase shifter can be configured to phase shift the audio signal 22 in the time domain to obtain a phase-shifted audio signal. In a short configuration, in which the phase shifter 26 is configured to phase shift the frequency representation 14 of the audio signal, its delayed version, respectively, the phase shifter can comprise a plurality of all-pass filters 28₁-28₁₆. In the example shown, all-pass filters 28₁-28₁₆are designed to accept detained parts 14'₁-14'₁₆. The term "all-pass filter" shall be understood to mean that the frequency range to be passed corresponds to the frequency range of the relevant part 14₁-14₁₆. In this case, the above may include examples in which each of the all-pass filters 28₁-28₁₆passes the full frequency range provided in the frequency representation, the passband of various all-pass filters 28₁-28₁₆may also differ from each other based on the different frequency bins contained in the corresponding parts 14₁-14₁₆.

Each of the all-pass filters 28 ₁ -28 ₁₆ is configured to phase shift the associated portion of the frequency representation of the audio signal.

Thus, the number of all-pass filter structures and/or the design of the all-pass filter structure may be the same, i.e. equal or comparable, or may, alternatively, differ between different all-pass filters 28 ₁ -28 ₁₆ .

The time delay provided by the delay lines 12 ₁ -12 ₁₆ may be the same or may differ for the different portions 14 ₁ -14 ₁₆ . As indicated in Fig. 2, portions of the frequency representation containing lower frequencies may be delayed with a higher time delay compared to portions of the frequency representation containing higher frequencies. From bin 1 to higher bins, the represented frequency may increase. As represented in the z-domain, the time delay may decrease with increasing frequencies.

Signals 32 ₁ -32 ₁₆ may contain the result of delay and phase shift, for example as the output of all-pass filters 28 ₁ -28 ₁₆ .

The envelope shaper 16 can be configured to receive signals 32 ₁ -32 ₁₆ and their unfiltered and non-delayed version, i.e. parts 14 ₁ -14 ₁₆ , i.e. a frequency representation of the audio signal 22. The parts 14 ₁ -14 ₁₆ can be understood as frequency sub-bands. The envelope shaper 16 can be configured to operate in the sub-band region. For example, the time resolution of the envelope shaper 16 can be at most or less than 4 milliseconds, such as 4 milliseconds, 3.5 milliseconds, 3 milliseconds or less.

The decorrelator 20 may comprise another transform module 35, which may provide an inverse operation compared to the transform module 24. For example, the transform module 34 may perform an inverse short-time Fourier transform (iSTFT). The frequency representation 18 of the combined form may comprise information regarding the frequency domain, which is present in each of the bins in such a way that the frequency representation 18 of the combined form can be processed according to the output of the transform module 24. Thus, the transform module 34 can receive processed versions of the parts 14 ₁ -14 ₁₆ of the frequency representation 14 and synthesize a synthesized signal 36 from the processed versions 14' ₁ -14' ₁₆ , for example, based on an overlap summation procedure. The signal 36 can be provided, for example, at the interface 38 of the decorrelator 20.

The envelope shaper 16 can be configured to shape the spectral bins in time and/or frequency. The shaping can be performed by means of the envelope shaper 26 for individual bins and/or for groups of bins, for example by implementing interdependent or at least group common shaping processing.

Referring again to the conversion module 24, it can be configured to receive and convert the audio signal 22 or a reverberated version thereof into a portion 14 ₁ -14 ₁₆ , wherein the number 16 is only an example. The reverberated version of the audio signal 22 can be an input in case the phase shifter 26 operates in the time domain and can therefore be located upstream of the conversion module 24. The conversion module 24 can perform a time-block discrete Fourier transform (DFT) or a short-time Fourier transform (STFT). The conversion module can be configured to convert blocks having an overlap of, for example, 50% within a tolerance range. For example, the tolerance range can be 0% to the maximum possible extent, at most 5%, at most 10%, at most 15% or more.

Blocks may contain a block length of, for example, 128 samples, 256 samples, or 512 samples, with 256 being preferred.

Fig. 3 shows a basic block diagram of the decorrelation 30. Compared with the decorrelator 20, the decorrelator 30 can additionally comprise a predelay 42, wherein the term "predelay" does not limit the delay to being implemented immediately before or after with respect to any particular block. The predelay 42 can be located in any stage before the envelope shaper 16, preferably also when operating in the frequency domain, after the transform module 24. Thus, for example, the sequence between the all-pass filters of the reverberation or phase shifter 26 and the predelay 42 can be rearranged compared to the illustration in Fig. 3. The predelay 42 or the delay block 42 can be designed with the possibility of additionally implementing the same and predetermined delay for a subset or all parts 14 ₁ -14 ₁₆ of the frequency representation. This may provide the ability to implement the same delay for each part 14 ₁ -14 ₁₆ or group thereof for combining processing in this stage, and also the ability to use delay lines 12 ₁ -12 ₁₆ to add, possibly, a separate delay so that it differs from the overall delay implemented in block 42. In one example, predelay 42 is configured to provide a constant predelay for all frequency bands of the spectrum.

Fig. 4 shows a basic block diagram of an all-pass filter 40 according to an embodiment, which can operate as at least part of one of the filters 28 ₁ -28 ₁₆ of the decorrelator 20 and/or 30. The all-pass filter 40 can comprise a Schroeder IIR filter structure, for example, and can comprise a forward branch 46 in combination with a reverse branch 48 in combination with a delay unit 52 to provide a corresponding output signal 54 based on an input signal 44 of the all-pass filter 40. The all-pass filter 28 of the decorrelator 20 and/or 30 can comprise one or more such all-pass filters 40 connected in series with each other. In order to provide different time delays in different all-pass filters 28 ₁ -28 ₁₆ , a different number of all-pass filter structures 14 can be connected in series.

In other words, Fig. 4 shows a cascade of all-pass filters.

Fig. 5 shows a basic block diagram of an all-pass filter structure 50, which is a nested all-pass filter structure. Alternatively or in addition to the all-pass filter structure 40, one or more all-pass filter structures 50 may form at least part of an all-pass filter 28 ₁ -28 ₁₆ of the decorrelator 20 and/or 30. Although two delay units 52 ₁ and 52 ₂ are shown, another and, in particular, a higher number of delay units 52 may be present, which as a result, possibly leads to an increased number of forward branches 46 and/or backward branches 48. Additionally, gains g ₁ /-g ₁ and/or g ₂ /-g ₂ may be applied.

When considering, for example, a series connection of the delay units 52 in one or more all-pass filter structures 40 and/or in one or more all-pass filter structures 50, different all-pass filters 28 ₁ -28 ₁₆ can be implemented in such a way that they contain a different time delay compared to other all-pass filters. For example, different delays of different all-pass filter structures and/or the designs of the all-pass filter structures can be based on a prime multiple of a local sampling frequency, for example, 48 kHz, used to obtain the frequency representation 14 of the audio signal 22. For example, a set of all-pass filter structures constituting at least a part of an all-pass filter can contain all-pass filter structures in the number of four, for example, the all-pass filter structure 40. The various delay units in them can be implemented with the ability to provide delays of 1, 2, 3 and 5. According to another example, the structures of all-pass filters in the number of four can provide delays of 1, 3, 5 and 7 units in the z-domain. These values can form a set of prime values, i.e. prime values in the number of 2, 3, 4, 5 or more can be grouped.

When translating, in this embodiment, sets of prime values, respectively, into possible operations of all-pass filters in the time domain, the time delays are based on a prime number that is a multiple of the inverse of the sampling frequency used to obtain the frequency representation of the audio signal in the embodiment. For example, the different time delays can be based on a prime number obtained by multiplying each of a set of prime numbers, as mentioned above, for example 1, 2, 3 and 5 or 1, 3, 5 and 7 by a downsampling factor used to form parts of the frequency representation of the audio signal, in order to obtain an intermediate result. Instead of an intermediate result, the next prime number relative to the intermediate result can be used. For example, given a downsampling factor of 128 and considering the prime numbers above, such a result could represent a delay of 131, 257, 383, and 641 on one side, and 131, 383, 641, and 907 on the other side, each delay being associated with a multiplication by 1 sample at the sampling rate, which, for a sampling rate of 48 kHz, is approximately 20.8 μs. Other prime numbers are possible without limitation.

Referring to, for example, Fig. 4, the gain factor g of the all-pass filter can be adapted to a value of 0.7 within a tolerance range of, for example, ±20%, ±10% or ±5%. However, the gain value can also have a negative value, for example, of -0.7 within the said tolerance range. Thus, the gain factor can be adapted to a value with an absolute value of 0.7 within the tolerance range.

In other words, in addition to the series-output pass configuration of Fig. 4, a nested configuration can also be implemented in which the delay element of the external all-pass Schroeder configuration is replaced by another internal all-pass configuration, or a combination of both configurations. Fig. 5 shows a simple nested cascade of all-pass filters.

Fig. 6 shows a basic block diagram of the decorrelator 60 according to an embodiment. The decorrelator 60 comprises a phase shifter 26, which is configured to operate in the time domain. The all-pass filter structure 28' may be configured to use the corresponding next primes compared to the sets of primes, as described in connection with the decorrelator 20 and/or 30. To ensure accurate operation of the decorrelator 60, it may comprise modules 24₁and 24₂transformations. While the module 24₁conversion can provide frequency representation of the audio signal, module 24₂transformations can receive a 22' reverberated or phase-shifted audio signal provided by a 28' phase shifter. The resulting parts are 14''₁-14''₁₆can be delayed by modules 12₁-12₁₆delays that achieve comparable input for the envelope shaper 16 compared to the decorrelator 20 and/or 30, while simultaneously providing the possibility of reverberation in the time domain. In this way, the frequency representation parts can form the frequency representation parts of the reverberated audio signal 22'.

According to embodiments, the decorrelator as described herein may be combined with additional functionality, i.e. the output signal may be further processed.

In other words, Fig. 6 shows an alternative implementation of the decorrelator relative to Fig. 2.

Additionally, the invented decorrelators can be combined with transient handling processing. Transients can cause artifacts in the decorrelated stereo signal, such as post-echo or unwanted panning effects. To mitigate this, transient handling can be combined with the decorrelator described in this document. Transient handling can dampen the output of the decorrelator to preserve the immediate onset waveform and suppress post-echo caused by the pre-delay.

Fig. 7 shows a basic block diagram of a decorrelator 70 according to an embodiment. The decorrelator 70 comprises at least a part of the decorrelator 10, whereby alternatively or additionally at least parts of the decorrelators 20, 30 and/or 60 can be arranged. The decorrelator 70 can comprise a signal processing stage 56 adapted to process the frequency representation 18 of the combined form or a signal based on it. The frequency representation 18 of the combined form can be considered as a mono signal, i.e. it can represent one channel. From the received mono signal, the processing stage can provide at least signals 58 ₁ and 58 ₂ representing a stereo signal.

The source pulling module 58, which models the perceptual effect of a spatially extended sound source from a mono signal of a point source and its decorrelated version, can be connected to the decorrelator 70. The source pulling module 58 can comprise filters 64 ₁ -64 ₂ , which provide the possibility of modeling the extension of the source based on a stereo signal having signals 58 ₁ and 58 ₂ . The modeling of the extension of the source can be performed, for example, in the frequency domain and can lead to output stereo signals 64 ₁ , for example, the left channel, and 64 ₂ , for example, the right channel. It should be noted that the source pulling module 58 can also form part of the decorrelator 70.

In other words, Fig. 7 shows a basic block diagram of the source extension provision processing.

Fig. 8 shows a basic block diagram of a processing system 80 according to an embodiment. The processing system 80 may comprise a decorrelator 10. Alternatively or additionally, a decorrelator 20, 30, 60 and/or 70 may be arranged. The processing system 80 comprises a processing stage 66 configured to transform a mid/side decomposed signal 68 into a left/right decomposed signal 72. Thus, the mid/side decomposed signal 68 may comprise at least a first signal 74 ₁ , for example representing one of a mid or a side portion, and a second signal 74 ₂ , representing the other portion. The processing stage 66 may be configured to transform signals 74 ₁ -74 ₂ and possibly additional signals into at least signals 76 ₁ -76 ₂ , representing a left channel and a right channel. One channel, for example the left channel L, can be obtained, for example, by summing the middle component M and the side component M+S; while the other, for example the right channel, can be obtained by subtracting one component from the other, for example MS. According to another approach, both channels can be obtained by using 50% or a factor of 0.5 for the said, i.e. 0.5(M+S) and 0.5(MS). Other factors and/or definition rules are possible.

According to an embodiment, a signal 74 ₁ is provided by a decorrelator of the processing system 80. Another signal 74 ₂ may be provided by a delay compensation module 78, which is connected in parallel with the decorrelator 10 and is configured to also receive the audio signal 22. The delay compensation module 78 is thereby connected to the processing stage 66. The delay compensation module 78 may be configured to provide a time delay that is comparable to the decorrelator. Preferably, for frequency domain embodiments, the delay is equal to the processing delay introduced by the STFT analysis/synthesis of the decorrelator. However, the decorrelator 10 may provide additional signal processing resulting in decorrelation such that the signal 74 ₂ may contain a similar delay compared to the signal 74 ₁ . According to an embodiment, the signal 74 ₂ may be unprocessed except for the time delay.

The decorrelator 10 in the processing system 80 may provide a frequency representation of the combined waveform as at least one part of the middle/side decomposed signal to the processing stage 66. The processing stage 66 may transform the frequency representation of the combined waveform together with the delay signal 74 ₂ into a left/right decomposed signal in the frequency domain. The output of the processing stage 66 may be the L/R signal 72. The decorrelator 10 itself may form a mono signal S (side, component 18), and in this respect it represents only a part of it. When handling transients, the direct part M (74 ₂ ; 74' ₂ ) and the output S of the decorrelator (signal 18) may become closely related, since the signal S will be damped and "replaced" by the amplified M signal (signal 74' ₂ ). As a consequence, both modules, the decorrelator and the "upmix module 66", are tightly coupled, and as a result, the processing stage 66 finally provides a decorrelated stereo signal. If the decorrelator is to operate autonomously with the mono output, for example, without the processing stage 66, then the direct signal with delay compensation, without any scaling, must be summed directly to the mono output to fill the blanked gap and provide the "full" signal.

In other words, Fig. 8 shows the decorrelator in M/S-to-L/R arrangement with mono input (average signal) delay compensation.

Fig. 9 shows a basic block diagram of a processing system 90 according to an embodiment. Compared to the processing system 80, the processing system 90 comprises a transient suppression module 82 configured to detect a transient in the audio signal 22 or its frequency representation 14 at the input of the decorrelator. The transient suppression module may comprise a transient detection module 84 configured to receive the audio signal 22 or its frequency representation. The transient detection module 84 may detect the transient in the audio signal, for example, by processing the audio signal 22. The transient suppression module 82 may further comprise a muting module 86 configured to receive the frequency representation 18 of the combined form and to mute it based on the control signal. However, it should be noted that the same or comparable effect can also be obtained by controlling the decorrelator 10 or the decorrelator contained in the processing system 90 to mute the output of the decorrelator. Thus, the muting module 86 can also form part of the decorrelator. However, the signal 74 ₁ forming the input of the processing stage 66 can be muted on the basis of a detected transient in the audio signal 22. The transient suppression module 82 can be configured to temporarily mute the part provided by the decorrelator to suppress echo signals in the processing stage 66, wherein the echo signals can be associated with pre-echo and/or post-echo. When operating in the time domain, a window can be used for soft muting in order to prevent additional transients that will be induced by muting. When performed in the frequency domain, the windowed STFT weighting described in connection with decorrelators 20, 30, and 60 can provide this effect automatically, i.e. synergistically.

With respect to the processing stage 66, the muting of the output of the decorrelator 10 may lead to an undesirable shift in the input energy of the signal processing stage 66. In order to prevent negative effects, the amplifier 82 may be connected between the delay compensation module 78 and the signal processing stage 66 to temporarily amplify the signal 74 ₂ in order to obtain an amplified signal 74' ₂ . The amplification of the signal 74 ₂ may be conditional with respect to the muting of the output of the decorrelator 10. Thus, the transient suppression module 82 may be configured to amplify a portion of the delay compensation module 78 according to the muting of a portion of the decorrelator.

The gain level may be fixed or controlled. According to one example, if used, the gain of amplifier 82 may be a factor of:

,

compared to the unmuffled part of the decorrelator. Thus, when the decorrelator output is muted, the amplifier 88 can amplify the signal 74 ₂ by:

,

when the signal is not amplified 74 ₂ during moments of time when the jamming is turned off, i.e. g=1.

Optionally, and also in order to avoid unwanted effects during the transient suppression, the transient suppression module 82 can be configured to suppress the detected transient in the audio signal and to suppress the next transient not earlier than a predetermined inhibit time. For example, the transient suppression module 82 can comprise a control module 92 configured to control and/or apply a hold time, a hysteresis and/or an inhibit time. For example, the hold time can be shorter than the inhibit time. The hold time can be related to the time during which the output of the decorrelator 10 is muted in response to the detected transient, i.e. a property determined by the transient detection module 84. The inhibit time can be longer than the hold time in order to avoid unwanted effects. For example, a hold counter, i.e. The time for jamming may be 1, 2, 4, 6, 7, or 8 blocks, although the ban time may be at least twice that time, such as at least 14, at least 20, at least 30, or 56 blocks, or any other time duration.

According to an example, the control module 92 can also provide a hysteresis to reduce the switching on/off of the transient suppression for audio signals, such as low-speed pulse trains. Thus, the inhibit time provided by the control module 92 can be a first inhibit time. The transient suppression module 82 can be configured to restart the inhibit time as a second inhibit time greater than the first inhibit time, in case a transient occurs during the first inhibit time. Thus, even if the hold time has expired, but the inhibit time has not yet expired, and in case a new transient is detected (regardless of whether the hold time has expired or not), the inhibit timer can be restarted. Optionally, the restarted inhibit timer can be longer compared to the canceled inhibit timer. In other words, when the very first transient is detected, then the hold counter and the inhibit counter are both started. The transient may be muted until the hold counter reaches its abort count, e.g. 8 blocks. Then the hold counter may be reset and the muting may be terminated. The inhibit counter may reach its abort count/be reset much later in time, e.g. after 56 blocks. If a new transient is detected during said ongoing inhibit counting process, only the inhibit counter is restarted, but with a higher abort count, e.g. 64 blocks. In this way, hysteresis is implemented by conditional switching and modifications of the abort count. Thus, during the inhibit counter operation, a new initiation of suppression or muting of the transient may be deactivated.

The transient suppression module 82 may be configured to operate in the frequency domain. Alternatively or additionally, the transient suppression module 82 may be configured to mute a portion of the decorrelator for a longer time than the pre-delay of the decorrelator. Thus, if a transient is detected in the audio signal 22, then the mute should still be in effect when the transient reaches the output of the decorrelator.

In other words, the decorrelators according to the embodiments operate in the short-time Fourier transform (STFT) domain on overlapping transform blocks of short duration. This provides a small processing delay of a few milliseconds, for example 2.7 milliseconds, assuming a transform size of 256 and a sampling frequency of 48 kHz, in contrast to the high delay of the PS/MDS decorrelator described in [2] or [3], which can achieve a delay time of 13.3 milliseconds at a sampling frequency of 48 kHz. In addition, the described decorrelators can be implemented using all-pass filters of very low computational complexity and can therefore be computationally much more efficient than the time-domain decorrelation described in [1] or [2]. If additional underlying spectral processing is needed or desired, such as source extension modeling, the described decorrelators can be interfaced directly to this processing cascade in the STFT domain to achieve low computational complexity.

The decorrelators described in this paper can therefore provide low processing latency and medium computational complexity. The decorrelators can be combined with additional underlying processing to model audio objects that have spatial dimensions, so-called spatially extended sound sources (SESS), with the perceptual property of "source extension".

In other words, Fig. 2 and Fig. 9 show preferred embodiments of the present invention. First, an input signal or audio signal (point source sound, for example) can be fed to a decorrelator 20 comprising a time-block DFT, for example with a block length of 256 samples and, for example, with 50% overlap. Next, the spectral bins of the DFT are delayed in time by a frequency-dependent duration, whereby low frequencies can have a higher delay and high frequencies can have a lower delay. For example, the delay can be 16 subband samples (42.7 milliseconds at 48 kHz) for low frequencies and can be reduced down to 1 subband sample for the highest bins, i.e. z ^-1 . The delay reduction over time can be linear, logarithmic, or otherwise rounded to integer numbers of subband samples. Then, each bin is sent through an all-pass filter, preferably comprising a chain of simple all-pass filters or a nested structure of all-pass filters. An exemplary all-pass filter is shown in Fig. 4. Another structure is shown in Fig. 5. With reference to Fig. 4, one possible chain may comprise or consist of four such all-pass filters. The parameter g may be chosen to be a component of, for example, 0.7, and the delays _Mi may be prime numbers. It should be noted that Fig. 4 shows the very first part of the chain, i.e. M ₁ . Since these filters may operate on downsampled frequency bands of the spectrum, for example with a downsampling factor of 128, the delays may be very low, for example, prime numbers 1, 2, 3, and 5, or, as another example, 1, 3, 5, and 7. Time/frequency envelope shaping may then be applied. The input signals to the envelope shaping may be the DFT bins themselves and their delayed and filtered versions. Finally, an overlap-sum IDFT may synthesize the output signal. The output signal may be further processed in the time domain to obtain a left/right stereo signal from the input mono signal in the configuration shown in Fig. 8. Alternatively, the left/right stereo signal may be collected in the frequency DFT domain and further processed in the frequency domain, e.g. for source extension/SESS modeling via fast convolution, if advantageous for overall computational efficiency. The configuration for source extension modeling is shown in Fig. 7. Unlike other embodiments, an alternative embodiment having delays _Mi may be selected as prime numbers approximately 128 times (according to the above-mentioned downsampling factor) larger than the prime numbers selected in the subband domain, for example 131, 257, 383 and 641 (for the set of prime values 1, 2, 3 and 5) or 131, 383, 641 and 907 (for the set of prime values 1, 3, 5 and 7). For different sets of prime values with different numbers of prime numbers and/or with different prime numbers, corresponding values may be selected. Additionally, the alternative embodiment may require an additional STFT to obtain a direct signal input to the time/frequency envelope shaper.

Fig. 9 shows an exemplary decorrelator in an M/S-to-L/R arrangement with processing for handling transients. Aspects of these embodiments are as follows:

- Transient detection detects the presence of an isolated transient.

- If a transient is detected, the decorrelated sound is muted during the "hold time" and the delay-compensated direct signal is amplified accordingly. To compensate for the coherent summation effect, a factor of 2/sqrt(2) is applied to amplify the direct signal, replacing the decorrelated signal.

- To prevent initiation in fast pulse trains that are perceived as tones, inhibition prevents initiation by the next transient for a certain "inhibition time"; the inhibition time is restarted by each new transient detection during the "hold time".

- Hysteresis prevents the transient detection from switching (e.g. by increasing the "inhibit time" in case of a re-initiated inhibition).

- Transient detection, muting, direct sound amplification, detection inhibition and hysteresis can be implemented predominantly in the STFT domain:

○ overlapping STFT blocks provides a smooth transition,

○ the damping time exceeds the decorrelator pre-delay,

○ jamming block counter to jam the decorrelated signal and amplify the direct signal,

○ Block inhibit counter to inhibit transient detection,

○ hysteresis to prevent switching when transients are detected.

Embodiments of the present invention relate to:

- a device/method for decorrelation of an audio signal,

- decorrelator, which includes:

○ DFT/IDFT pair (optional, when directly interfaced with SESS frequency domain processing),

○ subband delays; preferably low frequencies have higher delays and high frequencies have lower delays; delay distribution along frequency: linear, logarithmic, etc.,

○ subband all-pass filters; optional: low frequencies can have higher delay/order, and high frequencies have lower delay/order; higher-order all-pass filters can be realized by cascading lower-order all-pass filters,

▪ short IIR Schroeder filters in the (downsampled) subband DFT domain using primes with small integer delay in combination with frequency-dependent delays,

○ High-temporal-resolution (<4 ms) RF envelope control module operating in the subband domain; measures energy before and after delay/all-pass processing; controls the energy of the subband signal so that it matches (as closely as possible) the energy of the original subband signal,

- a low-latency decorrelator as part of the "source extent" modeling/processing (as opposed to a decorrelator based on the MPEG surround standard),

- interface with underlying source extent processing in the time or frequency DFT domain for computational efficiency,

- alternative implementation: all-pass filters before delays ("post-delays").

Fig. 10 shows a basic block diagram of a method 1000 according to an embodiment, which can be implemented, for example, by means of a decorrelator described in this document. The method 1000 comprises a step 1010, in which a plurality of parts are received, which are based on an audio signal. At 1020, each of the received parts is delayed to provide a plurality of delayed parts. 1030 comprises receiving and combining signals based on the delayed parts of the frequency representation. 1040 comprises receiving the frequency representation of the audio signal. 1050 comprises adjusting the energy of the delayed parts with respect to the frequency representation of the audio signal. 1060 comprises providing a frequency representation of a combined shape, for example, using an envelope shaper 16.

Although some aspects are described in the context of a device, it is obvious that these aspects also represent a description of the corresponding method, wherein the block or device corresponds to a step of the method or a feature of a step of the method. Similarly, aspects described in the context of a step of the method also represent a description of the corresponding block or element, or a feature of the corresponding device.

The invented encoded audio signal may be stored on a digital storage medium or may be transmitted over a transmission medium such as a wireless transmission medium or a wired transmission medium such as the Internet.

Depending on the specific implementation requirements, embodiments of the present invention may be implemented in hardware or software. The implementation may be performed using a digital storage medium, such as a floppy disk, DVD, CD, ROM, PROM, EPROM, EEPROM or flash memory, storing electronically readable control signals that interact (or are capable of interacting) with a programmable computer system such that the corresponding method is performed.

Some embodiments of the invention include a data carrier having electronically readable control signals that are capable of interacting with a programmable computer system such that one of the methods described herein is performed.

In general, embodiments of the present invention may be implemented as a computer program product with a program code, wherein the program code is configured to perform one of the methods when the computer program product is executed on a computer. The program code may, for example, be stored on a machine-readable medium.

Other embodiments comprise a computer program for implementing one of the methods described herein stored on a machine-readable medium.

In other words, an embodiment of the invented method is therefore a computer program having a program code for implementing one of the methods described in this document when the computer program is executed on a computer.

Accordingly, a further embodiment of the inventive methods is a data carrier (digital storage medium or machine-readable medium) containing a recorded computer program for implementing one of the methods described in this document.

Therefore, a further embodiment of the inventive method is a data stream or a sequence of signals representing a computer program for implementing one of the methods described in this document. The data stream or sequence of signals, for example, can be configured to be transmitted via a data connection, such as the Internet.

An additional embodiment comprises a processing means, such as a computer or a programmable logic device, configured to perform one of the methods described herein.

An additional embodiment comprises a computer having a computer program installed for implementing one of the methods described herein.

In some embodiments, a programmable logic device (e.g., a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor to perform one of the methods described herein. In general, the methods are preferably implemented by any hardware.

The above-described embodiments are merely illustrative of the principles of the present invention. It should be understood that modifications and changes to the arrangements and details described herein will be obvious to those skilled in the art. Therefore, they are intended to be limited only by the scope of the claims below and not by the specific details presented by way of description and explanation of the embodiments herein.

List of sources

[1] W. Oomen, E. Schuijers, B. den Brinker, and J. Breebaart, “Advances in Parametric Coding for High-Quality Audio,” paper 5852 (March 2003).

[2] J. Breebaart, S. van de Par, A. Kohlrausch and E. Schuijers, “High-quality Parametric Spatial Audio Coding at Low Bitrate,” paper 6072 (May 2004).

PS in QMF area:

[3] H. Purnhagen, J. Engdegard, J. Roden, and L. Liljeryd, “Synthetic Ambience in Parametric Stereo Coding,” paper 6074 (May 2004).

[4] J. Herre, K. Kjörling, J. Breebaart, C. Faller, S. Disch, H. Purnhagen, J. Koppens, J. Hilpert, J. Rödén, W. Oomen, K. Linzmeier and K.O. SE. Chong "MPEG Surround-The ISO/MPEG Standard for Efficient and Compatible Multichannel Audio Coding", J. Audio Eng. Soc., Vol. 56, No. 11, pp. 932-955 (November 2008).

Claims (58)

1. A decorrelator for an audio processing system, wherein the decorrelator comprises: - a plurality of delay modules (12), wherein each delay module (12) is configured to receive a portion (14 ₁ -14 _n ) of a frequency representation based on an audio signal (22); wherein each delay module (12) is configured to delay the received portion (14 ₁ -14 _n ) in order to provide a delayed portion (14' ₁ -14' _n ); and - an envelope former (16) configured to receive and combine signals based on delayed portions (14' ₁ -14' _n ) of a frequency representation; with the ability to receive a frequency representation of an audio signal (22); with the ability to regulate the energy of the delayed portions (14' ₁ -14' _n ) in relation to the frequency representation of the audio signal (22); and with the ability to provide a frequency representation of a combined form. 2. The decorrelator according to claim 1, wherein the different parts (14 ₁ -14 _n ) of the frequency representation contain the same or different numbers of frequency bins. 3. The decorrelator according to claim 1 or 2, additionally comprising a phase shifter (26) configured to phase shift the frequency representation (14) of the audio signal (22); or to phase shift the audio signal (22) in the time domain to obtain a phase-shifted audio signal (22). 4. The decorrelator according to claim 3, wherein the phase shifter (26) is configured to phase shift the frequency representation of the audio signal (22) and comprises a plurality of all-pass filters, wherein each all-pass filter (28) is configured to phase shift an associated portion (14 ₁ -14 _n ) of the frequency representation of the audio signal (22). 5. The decorrelator of claim 4, wherein the all-pass filter (28) of the plurality of all-pass filters comprises a set of all-pass filter structures (40; 50), such as Schroeder IIR filters, connected in series with each other; wherein the all-pass filter structures (40; 50) are configured to provide different time delays; or whereby the structures (40; 50) of all-pass filters contain a nested structure of all-pass filters. 6. The decorrelator according to claim 5, wherein the number of all-pass filter structures (40; 50) and/or the structure scheme of the all-pass filters are equal to or different between different all-pass filters (28). 7. A decorrelator according to claim 5 or 6, wherein the different time delays are based on a prime multiple of the local sampling frequency used to obtain the frequency representation of the audio signal (22). 8. The decorrelator according to one of claims 5-7, wherein the set of all-pass filter structures (40; 50) comprises four all-pass filter structures (40; 50) and they are configured to provide a delay of 1, 2, 3, and 5 or 1, 3, 5, and 7, respectively. 9. A decorrelator according to one of paragraphs 4-8, in which the gain of the all-pass filter (28) is adapted to a value with an absolute value of 0.7 within a tolerance range, for example 20%. 10. The decorrelator according to claim 3, wherein the phase shifter (26) is configured to phase shift the audio signal (22) in the time domain; wherein the phase shifter (26) comprises a set of all-pass filter structures (40; 50), such as Schroeder IIR filters, connected in series with each other; wherein the all-pass filter structures (40; 50) are configured to provide different time delays; or - in this case, the structures (40; 50) of all-pass filters contain a nested structure of all-pass filters. 11. The decorrelator of claim 10, wherein the various all-pass time delays are based on a prime multiple of the reciprocal of the sampling frequency used to obtain the frequency representation of the audio signal (22). 12. The decorrelator of claim 10 or 11, wherein the different time delays are based on a prime number obtained by multiplying each of a set of minimal prime numbers, such as 1, 2, 3 and 5; or 1, 3, 5 and 7, by a downsampling factor used to form parts (14 ₁ -14 _n ) of the frequency representation of the audio signal (22) to obtain an intermediate result; and for using the next prime number relative to the intermediate result, such as 131, 257, 383, 641 or 131, 383, 641, 907. 13. The decorrelator according to one of claims 10-12, comprising a first conversion module (24) for obtaining a frequency representation of the audio signal (22) from the audio signal (22) for the envelope generator (16); and comprising a second conversion module (34) for obtaining a frequency representation from the reverberated audio signal (22); wherein the parts (14 ₁ -14 _n ) of the frequency representation form parts (14 ₁ -14 _n ) of the frequency representation from the reverberated audio signal (22). 14. The decorrelator according to one of the preceding claims, wherein the parts (14 ₁ -14 _n ) of the frequency representation contain an equal or different number of frequency bins. 15. A decorrelator according to one of the preceding claims, configured to obtain parts (14 ₁ -14 _n ) of the frequency representation in the number 16. 16. A decorrelator according to one of the preceding claims, configured to obtain a frequency representation with frequency bins in the number 128 or 129. 17. A decorrelator according to one of the preceding claims, wherein the decorrelator is configured to additionally implement the same and predetermined delay for a subset or all parts (14 ₁ -14 _n ) of the frequency representation. 18. A decorrelator according to one of the preceding claims, wherein the delay modules (12) associated with the spectral portion (14 ₁ -14 _n ) of the plurality of delay modules (12) are configured to delay the associated portion (14 ₁ -14 _n ) of the frequency representation differently compared to the delay modules (12) associated with other spectral portions (14 ₁ -14 _n ). 19. A decorrelator according to one of the preceding claims, wherein the plurality of delay modules (12) are configured to delay portions (14 ₁ -14 _n ) of the frequency representation containing lower frequencies with a higher time delay compared to portions (14 ₁ -14 _n ) of the frequency representation containing higher frequencies. 20. The decorrelator of claim 19, wherein the relationship between the different time delays is one of linear, logarithmic, and/or based on subband sample rounding. 21. A decorrelator according to one of the preceding claims, comprising a transform module (24) for receiving and transforming an audio signal (22) or a reverberated version of the audio signal (22) into parts (14 ₁ -14 _n ) by performing a time-block discrete Fourier transform (DFT) or a short-time Fourier transform (STFT); wherein the transform module (24) is configured to transform blocks having an overlap of 50% within a tolerance range. 22. A decorrelator according to one of the preceding claims, comprising a transform module (24) for receiving and transforming the audio signal (22) or a reverberated version of the audio signal (22) into parts (14 ₁ -14 _n ) by performing a time-block discrete Fourier transform (DFT) or short-time Fourier transform (STFT); wherein the blocks comprise a block length of 256 samples. 23. A decorrelator according to one of the preceding claims, comprising an inverse transform module (34) for receiving processed versions of portions of the frequency representation (14) and for synthesizing a synthesized signal from the processed versions based on an overlap summation procedure. 24. A decorrelator according to one of the preceding claims, in which the envelope shaper (16) is designed to operate in the subband region and with a time resolution of less than 4 ms. 25. A decorrelator according to one of the preceding claims, comprising an interface (38) for providing a signal (36) based on a frequency representation of the combined form. 26. A decorrelator according to one of the preceding claims, wherein the envelope shaper (16) is to shape the spectral bins in time and/or frequency individually or as a group, for example by implementing interdependent or at least group common shaping processing. 27. A decorrelator according to one of the preceding claims, comprising a signal processing stage (66) adapted to receive a signal based on a frequency representation of a combined form as a mono signal and to process the mono signal into at least a stereo signal. 28. A decorrelator according to one of the preceding claims, comprising a signal processing stage (66) adapted to process a frequency representation of the combined form into at least a stereo audio signal; and adapted to model the extent of the source based on at least the stereo signal, for example in the frequency domain. 29. An audio processing system comprising: - a decorrelator according to one of the preceding points; and - a processing cascade (66) for converting the middle/side decomposed signal into a left/right decomposed signal. 30. The processing system according to claim 29, in which one part (74 ₁ ) of the middle/side decomposed signal is provided by means of a decorrelator, and the other part (74 ₂ ) is provided by means of a delay compensation module (78) connected in parallel with the decorrelator and connected to the processing cascade (66). 31. The processing system according to claim 30, comprising a transient suppression module (82) configured to detect a transient in the audio signal (22) or in its frequency representation (14) in the input of the decorrelator; - wherein the transient suppression module (82) is designed with the possibility of temporarily silencing the part (74 ₁ ) provided by means of the decorrelator in order to suppress echo signals in the processing cascade. 32. The processing system according to claim 31, in which the transient suppression module (82) is configured to amplify a portion of the delay compensation module corresponding to the damping of a portion of the decorrelator. 33. The processing system according to claim 32, in which the transient suppression module (82) is configured to amplify a portion of the delay compensation module by a factor of: , compared to the undamped part of the decorrelator. 34. The processing system according to one of paragraphs 31-33, in which the transient suppression module (82) is configured to suppress the detected transient and to suppress the next transient no earlier than a predetermined inhibition time. 35. The processing system according to one of claims 31-34, wherein the inhibition time is a first inhibition time; wherein the transient suppression module (82) is configured to restart the inhibition time as a second inhibition time exceeding the first inhibition time if a transient occurs during the first inhibition time. 36. A processing system according to one of paragraphs 31-35, in which the transient suppression module (82) is designed with the ability to operate in the frequency domain. 37. The processing system according to one of paragraphs 31-36, in which the transient suppression module (82) is designed with the possibility of suppressing a part of the decorrelator for a longer time compared to the pre-delay of the decorrelator. 38. The processing system of any one of claims 29-37, wherein the decorrelator is to provide a frequency representation of the combined waveform as part of a mid/side decomposed signal to a processing stage; and the processing stage is to convert the frequency representation of the combined waveform and a delayed version of the audio signal (22) into a left/right decomposed signal in the frequency domain. 39. A method for processing an audio signal, comprising the steps of: - receive (1010) a plurality of parts of a frequency representation based on an audio signal; - delay (1020) each of the received parts in order to provide a plurality of delayed parts; and - receive (1030) and combine signals based on delayed parts of the frequency representation; - receive (1040) frequency representation of the audio signal; - regulate (1050) the energy of the delayed parts in relation to the frequency representation of the audio signal; and - provide (1060) frequency representation of the combined form. 40. The method according to claim 39, further comprising the steps of: - detect a transient in an audio signal (22) or in its frequency representation (14); - temporarily mute the part (74 ₁ ) provided by the decorrelator in order to suppress echo signals in the processing cascade. 41. A machine-readable medium on which a computer program is stored for implementing, when executed on a computer or processor, the method according to paragraph 39 or 40.