Method and unit for performing dynamic range control

Abstract

This document describes a dynamic range control unit (210) configured to apply a dynamic range control, referred to as DRC, to an audio signal (211). The DRC unit (210) is configured to: downsample a subband signal (212) derived from the audio signal (211) to provide a downsampled subband signal (321); determine a DRC gain (329) based on the downsampled subband signal (321); and apply the DRC gain (329) to the subband signal (212) to provide a compressed subband signal (213) of a compressed audio signal (214).

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to the following prior applications: U.S. Provisional Application 63 / 027,517 (reference number: D17089USP1), filed May 20, 2020, and European Application 20175651.7 (reference number: D17089EP), filed May 20, 2020, which are incorporated herein by reference. Technical Field

[0003] This document relates to a method and corresponding device or unit for applying dynamic range control (DRC) to audio signals, especially multi-channel audio signals. Background Technology

[0004] Dynamic range control, or dynamic range compression (DRC), can be used in various applications, such as to allow audio content to be presented with high perceived quality in noisy environments. Specifically, the audio signal is modified in a way that makes the (compressed) audio signal robust to ambient noise (and therefore intelligible and / or audible). DRC achieves this by increasing the density of the audio signal in a dynamical aspect. DRC can be applied to increase the average power of the audio signal without increasing the peak level of the audio signal. By doing so, an increased signal-to-noise ratio (SNR) can be achieved.

[0005] Applying DRC to multi-channel audio signals may be associated with high computational complexity. Furthermore, applying DRC to audio signals can result in audible artifacts, which may be unpleasant for the average listener.

[0006] This document addresses the technical challenges of implementing DRC with relatively low computational complexity and / or relatively high perceptual quality, especially in the context of multi-channel and / or multi-object coding systems such as Dolby AC-4. Summary of the Invention

[0007] According to one aspect, a dynamic range control unit is described, configured to apply dynamic range control, referred to as DRC, to an audio signal. The DRC unit is configured to downsample a sub-band signal derived from the audio signal to provide a downsampled sub-band signal. Furthermore, the DRC unit is configured to determine a DRC gain based on the downsampled sub-band signal and to apply the DRC gain to the sub-band signal (i.e., before downsampling it) to provide a compressed sub-band signal of the compressed audio signal.

[0008] It should be noted that the downsampling of the subband signal is performed after conversion to the subband domain (e.g., QMF filter bank processing). Therefore, the downsampling of this invention is... In addition toany down-sampling inherent in the sub-band domain conversion In addition to Such down-sampling inherently included in the sub-band domain conversion is for example discussed in US 2007 / 0078645, wherein the down-sampling is inherently implemented in the filter bank converting the time domain signal into the decimated sub-band signal.

[0009] Further, it is noted that according to the present application, the determination of the DRC gain is only performed on the down-sampled sub-band signal. The DRC gain is then applied to the sub-band signal output from the sub-band domain conversion, and not to an additionally down-sampled version thereof. This is fundamentally different from prior art solutions like US 2007 / 0078645, wherein the down-sampled sub-band signal is used for all downstream audio processing.

[0010] According to another aspect, a decoding unit is described, wherein the decoding unit is configured to derive a dynamic range compressed audio signal from a bitstream indicative of the audio signal. The decoding unit comprises a DRC unit as described herein.

[0011] According to another aspect, a method for applying a dynamic range control, referred to as DRC, to an audio signal is described. The method comprises down-sampling a sub-band signal derived from the audio signal to provide a down-sampled sub-band signal. Further, the method comprises determining a DRC gain based on the down-sampled sub-band signal, and applying the DRC gain to the sub-band signal to provide a compressed sub-band signal of a compressed audio signal.

[0012] According to another aspect, a software program is described. The software program can be adapted for execution on a processor and for performing the method steps outlined in the present document when carried out by the processor.

[0013] According to another aspect, a storage medium is described. The storage medium can comprise a software program adapted for execution on a processor and for performing the method steps outlined in the present document when carried out by the processor.

[0014] According to another aspect, a computer program product is described. The computer program can comprise executable instructions for performing the method steps outlined in the present document when carried out by a computer.

[0015] It should be noted that the methods and systems as outlined in the present patent application - including preferred embodiments thereof - can be used stand-alone or in combination with other methods and systems disclosed in the present document. Further, all aspects of the methods and systems outlined in the present patent application can be combined in any combination. In particular, features of the claims can be combined with each other in any manner. BRIEF DESCRIPTION OF DRAWINGS

[0016] The application is explained below with reference to the drawings in an exemplary manner, in which: - Fig. 1 shows a schematic block diagram of a DRC system according to an embodiment of the present application; and - Fig. 2 shows a schematic block diagram of a DRC system according to another embodiment of the present application.

[0017] Figure 1 a block diagram of an example encoding system is shown;

[0018] Figure 2 an example scheme of applying DRC in a subband domain is illustrated;

[0019] Figure 3 a block diagram of an example DRC unit is shown; and

[0020] Figure 4 a flowchart of an example method for applying DRC to an audio signal is shown. DETAILED DESCRIPTION

[0021] As outlined above, the present document addresses the technical problem of implementing dynamic range control at low computational complexity and / or high perceptual quality. In this context, Figure 1 A block diagram of an example audio encoding system (also referred to as codec) 100 is shown, which comprises an encoding unit 110 (also referred to as encoder) and a corresponding decoding unit 120 (also referred to as decoder). The encoding unit 110 is configured to generate a bitstream 112 of the (multi-channel) audio signal 111, wherein the bitstream 112 is indicative of encoded audio data and metadata of the audio signal 111. The decoding unit 120 is configured to generate a reconstructed (multi-channel) audio signal 113 based on the bitstream 112. The decoding unit 120 can comprise a DRC unit as described in the present document.

[0022] In order to enable the decoding unit 120 to perform DRC on the reconstructed audio signal 113, the encoding unit 110 can be configured to provide DRC profile data (also referred to as DRC metadata) as part of the bitstream 112, which can be used by the decoding unit 120 to perform DRC. In particular, the encoding unit 110 can use the profile data to steer or control DRC in the decoding unit 120, in particular in a DRC unit of the decoding unit 120. Thus, the profile data can comprise control data for performing DRC in the decoding unit 120. The profile data can be indicative of a compression curve to be used for DRC.

[0023] DRC can be performed in a subband domain (also referred to as transform domain). For this purpose, the audio signal 111 can be transformed or converted from the time domain to the subband domain using a forward transform or analysis filter bank, like an analysis (pseudo) quadrature mirror filter (QMF) bank. After performing DRC in the subband domain, the dynamic range compressed subband signals can be transformed or converted back from the subband domain to the time domain using an inverse transform or synthesis filter bank, like a synthesis QMF bank.

[0024] Figure 2The illustration shows an example scheme for performing DRC in the subband domain. The audio signal 211 (e.g., an analysis filter bank 201) can be used for forward transformation or analysis of the filter bank 201 (e.g., analysis of a pseudo-QMF bank). Figure 1 The audio signal 111) is transformed or converted from the time domain to the subband domain. As a result of the forward transformation or analysis, multiple subband signals 212 corresponding to multiple subbands or frequency bins are obtained. For example, K subband signals 212 can be obtained, such as K = 64.

[0025] DRC can be applied to different subband signals 212 in DRC unit 210, wherein DRC unit 210 may take into account profile data 215 (which can be provided by encoding unit 110 in bit stream 112). As mentioned above, profile data 215 may also be referred to as DRC metadata.

[0026] DRC unit 210 can generate multiple (dynamic range) compressed sub-band signals 213 (specifically K compressed sub-band signals 213) from corresponding multiple sub-band signals 212. The compressed sub-band signals 213 can be transformed or converted to the time domain using inverse transform or synthesis filter bank 203 (e.g., synthesis of pseudo-QMF banks) to provide a (dynamic range) compressed audio signal 214. It should be noted that in this document, the term "compression" refers to dynamic range compression (not data compression).

[0027] The following describes an example DRC scheme for a pseudo-QMF group, where K = 64 subband signals 2^12. It should be noted that the aspects described below are applicable in a similar manner to other time-domain / frequency-domain or subband-domain transforms or other analysis / synthesis filter banks.

[0028] The input data 212 fed to the DRC unit 210 can be organized as a K×N matrix, where K represents the number of subbands and N is the number of time slots for joint processing (e.g., N = 32). The N subband samples 212 corresponding to the N time slots can be referred to as blocks or frames. The output data 213 from the DRC unit 210 typically has the same dimensions as the input data 212.

[0029] DRC unit 210 can operate in Q DRC bands (e.g., Q=4), each DRC band covering one or more sub-bands or frequency blocks. Therefore, K sub-bands can be grouped into Q bands. Various sub-band or frequency block grouping methods can be used. For example, the following non-uniform sub-band grouping (also known as banding) can be used:

[0030] Q = 4 frequency bands; and

[0031] start_band = [1,2,4,19,K] max +1].

[0032] In other words, the first frequency band may include frequency compartment k=1, the second frequency band may include frequency compartments k=2, 3, the third frequency band may include frequency compartments k=4, 5, ..., 18, and the fourth frequency band may include frequency compartments k=19, 20, ..., K. max To reduce computational complexity, when calculating the DRC gain used to perform DRC, only K of the K subband signals 212 need to be considered. max Each sub-band signal, where K max <K。

[0033] Figure 3 A block diagram of an example DRC unit 210 is shown. Each sub-unit is described below. The DRC unit 210 may be included in... Figure 3 Any subset of the subunits described in the context.

[0034] DRC unit 201 may include downsampling unit 301, which is configured to sample from K max Each sub-band signal 212 generates K max A downsampled subband signal 221. Specifically, the temporal resolution of subband signal 212 can be reduced by a downsampling factor M. Therefore, we can apply K. max The ×N / M input data matrix performs gain calculations (for each channel of the multi-channel audio signal 211, for example, for two channels in the case of the stereo signal 211, thus generating an input data tensor). Therefore, the dimension of the input data used to calculate one or more DRC gains can be reduced by a downsampling factor M.

[0035] It should be noted that the downsampling performed by factor M as described above is separate from any downsampling that may be inherent in the conversion to the subband domain. For example, a QMF filter bank can be configured to provide downsampling of the subband signal. The downsampling in downsampling unit 301 is performed in addition to any such downsampling.

[0036] In a preferred example, the downsampling factor M is chosen such that N is an integer multiple of M to allow for uniform sampling. Furthermore, considering the temporal resolution of human hearing—that is, assuming the new sampling period is less than (or equal to) the integration time of the human ear—downsampling can be performed without using a low-pass filter, further reducing computational complexity. Therefore, subband signal 212 can be downsampled efficiently by using the downsampling factor M, retaining only every Mth sample or time slot of subband signal 212.

[0037] Given that using complex pseudo-QMF groups can lead to oversampling with a sampling factor of 2, performing downsampling with a factor of M=2 usually has no perceptible impact on the quality after applying DRC.

[0038] As a result of downsampling the subband signal 212 by pressing the sampling factor M, this K max The downsampled subband signal 321 represents the number of time slots, which is a factor M that is less than the number of time slots of the corresponding subband signal 212. The time slots of the downsampled subband signal 321 can be identified using an index n.

[0039] Loudness unit 302 can be configured to determine a loudness value or loudness estimate 322 (in the linear domain) for each sub-band group (or DRC band) i = 1, 2, ..., Q, for each time slot n = 1, 2, ..., N / M, and possibly for each channel l = 1, 2, ..., L (e.g., L = 2) of the (multi-channel) audio signal 211.

[0040] For DRC band i and time slot n, loudness can be calculated as follows. First, the in-band power P of each channel l (e.g., l∈{1,2}) inl It can be determined according to the following formula

[0041]

[0042] Among them, start_band i It is the first (lowest) frequency band in the i-th DRC band, and where, start_band i+1 -1 is the last (highest) frequency band in the i-th DRC band. It is sample x knl The real part of (for time slot n, frequency bin k, and channel l), and It is sample x knl The imaginary part.

[0043] Subsequently, the in-band power P can be adjusted. inl Find the square root to calculate the in-band power value P. inl Convert to (in-band) amplitude value, i.e.

[0044]

[0045] Finally, the loudness index (α = 0.67) can be used to apply Stevens' power law to determine the loudness value or loudness estimate L. inl 322 (for DRC band i, time slot n, and channel l), i.e.

[0046]

[0047] In this way, Q×L loudness values ​​or loudness estimates L are obtained at each time point or time slot n. inl322 (in the linear domain), a loudness value or loudness estimate for a combination of each DRC band i (i = 1, 2, ..., Q) and each channel l (l = 1, 2, ..., L).

[0048] The linear-to-logarithmic converter unit 303 can be configured to convert the loudness value L inl The loudness value 322 is converted from the linear domain to the logarithmic domain. The DRC typically uses a static compression curve, which is part of the DRC profile. Since compression curves are generally defined in the logarithmic domain, the loudness value 322 in the linear domain can be converted to decibels using the following formula.

[0049]

[0050] This provides a logarithmic loudness value or loudness estimate. 323.

[0051] Loudness-to-gain mapping unit 304 can be configured to be based on the corresponding loudness value or loudness estimate. 323 is used to obtain a gain value of 324 (referred to herein as the mapped gain value). For this purpose, a compression curve can be used. The parameterization of the compression curve can be defined, for example, in the ETSI standard ETSI TS 103 190 (specifically for DRC in AC-4). This document is incorporated herein by reference in its entirety.

[0052] Compression curves can be piecewise linear functions in the logarithmic field, including at least some of the following seven subfunctions:

[0053]

[0054] The parameters of the compression curve F(L) can be provided as configuration data 215. Example parameters are different (linear) gain values ​​G for different loudness ranges. maxcut G sectioncut G sectionboost and / or G maxboost .

[0055] Other example parameters are thresholds used to define different segments of the compression curve.

[0056] ·L maxcut Used to define the start (using gain value G) maxcut Perform the loudness value or loudness estimate of maximum downward compression;

[0057] ·L sectioncut and L 0high , used to define one or more loudness values ​​or loudness estimates at which intermediate downward compression begins;

[0058] ·L 0low and L0high This is used to define loudness values ​​or loudness estimation intervals that are not compressed;

[0059] ·L sectionboost and L 0low Used to define one or more loudness values ​​or loudness estimates at which intermediate upward compression begins; and / or

[0060] ·L maxboost Used to define the start (using gain value G) maxboost Perform the loudness value or loudness estimate of the maximum upward compression.

[0061] The compression curve F(L) is configured to map the loudness value or loudness estimate L 323 to the corresponding (mapped) gain value 324. The (mapped) gain value 324 can be determined for each time slot n, each DRC band i, and / or each channel l.

[0062] The segmented function F(L) comprises a lift segment (upward compression or "expansion") centered on a reference value and a cut segment (downward compression or simply "compression"). The reference value may correspond to the average dialogue level within the audio signal 211, also known as the dialogue normalization. In the above formula, the reference value is assumed to be 0 dB. Alternatively, a reference value based on the dialogue normalization value can be used (which may be indicated in bitstream 112).

[0063] The operating point O of the compression curve F(L) inl It can be determined by the loudness deviation ΔL between the instantaneous broadband loudness level and a reference level (e.g., dialogue normalization). Therefore, the (mapped) gain value F for each DRC band i is... inl 334 (for each time slot n and / or channel l) can be given by the following formula.

[0064]

[0065] Among them, O inl It is the operating point 331 of the compression curve, which depends on the loudness deviation ΔL. This refers to the in-band loudness value or loudness estimate in the logarithmic domain mentioned above. The calculation bias ΔL and operating point O are also relevant. inl Further details regarding the preferred method are provided below.

[0066] The stereo or multi-channel link unit 305 can be configured to determine a single gain value 325 for a subgroup or all channels of the audio signal 211. Typically, it is desirable to preserve the multi-channel (especially stereo) picture of the audio signal 211. This can be achieved by applying the same gain value to all channels l (l = 1, 2, ..., L) of the audio signal 211. Specifically, for all channels l (l = 1, 2, ..., L) of the audio signal 211, the minimum gain value (i.e., the maximum attenuation under compression and the minimum amplification under expansion) can be selected as the common gain value 325. Therefore, in the case of a dual-channel audio signal 211, the gain value 325 can be selected as...

[0067] F in =min(F in1 F in2 ).

[0068] This principle can be extended to a larger number (L) of channels.

[0069] Gain smoothing unit 306 can be configured to smooth (map) gain values ​​324 and 325 over time. Specifically, exponential smoothing can be applied to (map) gain values ​​324 and 325 at consecutive moments to avoid sudden changes in the output signal due to compression or expansion. Smoothing gain G in 326 can be identified as

[0070] G in =αF in +(1-α)·G i，n-1 ,

[0071] Among them, G i,n-1 It is the smoothing gain of the previous time slot n-1, 326, where F in These are the (instantaneous or mapped) gain values ​​of 324 and 325 calculated for the current time slot n, where G... in It is the smoothing gain of the current time slot n, 326.

[0072] The smoothing factor α can depend on the (in-band) loudness level of the signal or the loudness estimate. 323 (for one or more channels of audio signal 211) is below or above the reference point or operating point 331, and the loudness level or loudness estimate. Whether 323 is rising or falling needs to be determined. Specifically,

[0073]

[0074] Among them, respectively

[0075]

[0076] and

[0077]

[0078] The parameter α for performing gain smoothing can be provided in configuration file data 215. attack and α release .

[0079] Furthermore, two distinct smoothing factors can be distinguished based on the loudness jump between two consecutive moments or time slots n (i.e., the difference between loudness estimates 323). For relatively large jumps, a more rapid response from the DRC is generally expected. For relatively small changes, the response from the DRC should generally be gradual. For this purpose, two thresholds are anticipated in the aforementioned ETSI standard: T attack and T release (These two thresholds can be indicated within configuration file data 215). Therefore, the smoothing factor α can be determined as follows:

[0080]

[0081] and

[0082]

[0083] The relationship between the smoothing factor and the corresponding time constant can be:

[0084]

[0085] Among them, T s T is the sampling period after downsampling, where τ is the time constant. s Both τ and τ are typically expressed in milliseconds. Parameter T s And τ can be indicated in configuration file data 215.

[0086] Gain compensation unit 307 can be configured to add a constant offset to smoothing gain value 326 to provide a modified gain value 327. Metadata associated with (AC-4) bitstream 112 can indicate the average dialogue level (i.e., dialogue normalization) of audio signals 111, 211 before DRC application. Furthermore, bitstream 112 can indicate the desired output reference level after DRC application. The difference B between the two levels can be added as a constant gain B 332 to smoothing gain 326 to generate a control signal or control gain 327, i.e.

[0087] C in =G in +B.

[0088] The logarithmic-to-linear converter unit 308 can be configured to convert the control signal or control gain 327 from the logarithmic domain to the linear domain, so that the control gain 327 can be combined with the sub-band signal 212. This can be accomplished by the following inverse operation.

[0089]

[0090] This provides a control gain of 328 in the linear domain (for time slot n and DRC band i).

[0091] Upsampling unit 309 can be configured to upsample control gain 328 by a factor M to provide upsampled control gain 328 for each time slot of subband signal 212. Therefore, the control signal or gain 328 can achieve the sampling rate of subband signal 212. This can be achieved efficiently by separating the original sample of control gain 328 with M-1 zeros and filling these zeros with the original sample, similar to a "sample and hold" operation.

[0092] c i，j+n·M =c in j = 0, 1, ..., M-1.

[0093] Gain application unit 310 can be configured to apply the (upsampling) control signal or gain 329 to all channels of subband signal 212 via multiplication:

[0094]

[0095] Subband signals 212 of the same DRC band i (i.e., samples of different subband signals 212) can use the gain c of that DRC band i. in 329 Compression. As a result, a compressed subband signal 213 is obtained. The compressed subband signal 213 can be transformed or converted to the time domain using a subband-to-time domain (inverse) transform or a synthesis filter bank (e.g., a synthesis pseudo-QMF bank) to determine the compressed audio signal 214.

[0096] To reiterate, the control gain in unit 310 is applied to the subband signal converted from the subband domain, not its downsampled version. Therefore, the downsampling in unit 301 has only a computational impact on the determination of the DRC gain, not its application.

[0097] As described above, the operating point O of the compression curve F(L) can be determined individually for each DRC frequency band i and time slot n. in 331. The deviation ΔL between the instantaneous broadband loudness level and the average dialogue level across the entire audio signal 211 can be used as a basis. t The estimated value is used to determine the operating point 331. The average dialogue level is typically part of the metadata of the audio signal 211 provided in bitstream 112.

[0098] To determine the operating point 331, the block average (BA) power value in the time direction of the entire input data block (distributed across N time slots n = 1, 2, ..., N) can be determined. For simplicity, channel index l is omitted below. The BA power value can be determined as follows:

[0099]

[0100] This will produce in-band BA values ​​for different DRC frequency bands i (i = 1, 2, ..., Q). These BA values The loudness value can be converted to decibels using Stevens' power law and a common logarithm (base 10). As outlined above for units 302 and 303, the transformed BA values ​​can be exponentially smoothed to form a short-term average (STA).

[0101]

[0102] Additionally, STA can be smoothed to produce a long-term average (LTA).

[0103]

[0104] Here, t is the time index indicating a consecutive block of input data. Therefore, the in-band BA value can be determined for consecutive blocks of data or frames. The short-term average (STA) and long-term average (LTA).

[0105] LTA This can represent the in-band instantaneous loudness level (of the data block at time t). Instantaneous broadband loudness level (Across all DRC bands) the value can be obtained from the narrowband LTA by converting the LTA to a power value.

[0106]

[0107] Sum these power values ​​for all DRC bands.

[0108]

[0109] The sum is then converted to the corresponding loudness value in decibels (i.e., in the logarithmic domain):

[0110]

[0111] Loudness level The instantaneous deviation from a reference level (e.g., dialogue normalization) can then be determined as the difference.

[0112]

[0113] The operating point O in different DRC frequency bands i (i = 1, 2, …, Q) in 331 can be obtained by shifting the narrowband loudness level a deviation ΔL proportional to the distribution of the broadband power across different DRC frequency bands t for calculation as follows:

[0114]

[0115] For all time slots n, in the t-th block of the input data. The operating point 331 can thus be considered as the in-band counterpart of the average conversation level in the audio signal 211.

[0116] Both STA and LTA generally require initial values for initialization at time t = 0. In the example, at the start (i.e., at time t = 1), the values and can be initialized with the BA value for initialization.

[0117] Whenever the absolute difference between any current LTA value (at time t) in decibels and the previous LTA value (at time t - 1) exceeds a predetermined threshold (e.g., the threshold is 10 dB), the running STA and LTA values can be reset to the instantaneous BA value This enables the DRC to quickly respond to sudden scene changes. Whenever a scene change is detected, the corresponding DRC gain can be set to zero.

[0118] As described above, decimation in time in the QMF or sub-band domain can be used to reduce the computational complexity of the DRC. Alternatively or additionally, the application of the DRC may be restricted to a certain frequency, such as 16 kHz. In particular, an upper cut-off frequency below the Nyquist frequency can be used to reduce the number of QMF bands in the highest DRC frequency band. For example, only K max (<K) sub-bands are considered. This can reduce the number of MAC (multiply-accumulate) operations in the loudness detector unit 302.

[0119] The power value of each DRC frequency band i (i = 1, 2, …, Q) is typically accessed on multiple occasions. The power value can be calculated and stored once, and then can be reused via a lookup operation when needed.

[0120] To further reduce computational complexity, channel linking can be performed early (e.g., in loudness detector unit 302). For example, it can be determined which channel to maintain based on the power values ​​in different channels (for each DRC band i, i = 1, 2, ..., Q) to determine the joint control gain 329 for all channels of the audio signal 211. By doing so, computational complexity can be reduced.

[0121] This document proposes using Stevens' power law to approximate loudness. Alternative methods, such as A-weighted or K-weighted methods, can be used to approximate loudness. However, considering computational complexity, using Stevens' power law is preferred because it can be implemented efficiently, especially when combined with the lin2log operation (i.e., combining the aforementioned transformation from the linear domain to the logarithmic domain, for example, using the 20log operation). 10 (·)).

[0122] Figure 4 A flowchart is shown of an example (possibly computer-implemented) method 400 for applying dynamic range control or dynamic range compression (DRC) to an audio signal 211. The audio signal 211 may include multiple channels l, l = 1, 2, ..., L, where L ≥ 2. In other words, the audio signal 211 may be a multi-channel audio signal.

[0123] Method 400 includes downsampling 401 of a sub-band signal 212 obtained from audio signal 211 to provide a downsampled sub-band signal 321. Downsampling 401 can be performed by a downsampling factor M, as outlined in this document. Method 400 can be performed at a decoding unit 120 configured to decode a bitstream 112, wherein the bitstream 112 indicates audio signal 211. One or more sub-band signals 212 can be obtained from audio signal 211 using a time-frequency (forward) transform or analysis filter bank (such as an analysis pseudo-QMF group). In other words, method 400 can include applying a forward transform or analysis filter bank to audio signal 211 to determine one or more (especially K) sub-band signals 212.

[0124] In this way, K sub-band signals 212 with K different sub-bands or frequency bins can be provided, wherein the K sub-band signals 212 can be recombined using inverse time-frequency transformation or synthesis filter banks (e.g., synthesis pseudo-QMF banks) to provide audio signal 211.

[0125] Downsampling 401 of one or more subband signals 212 can be performed in an efficient manner by selecting only the Mth sample from the corresponding subband signal 212 to provide the corresponding downsampled subband signal 321.

[0126] Method 400 further includes determining 402 a DRC gain 329 based on the downsampled subband signal 321. The DRC gain 319 (also referred to herein as the control gain) can be determined using the scheme outlined in Figure 3 the context of.

[0127] Additionally, method 400 includes applying 403 the DRC gain 329 to the subband signal 212 to provide a dynamically range compressed subband signal 213 of the dynamically range compressed audio signal 214. Thus, the DRC gain 329 that has been determined using the downsampled subband signal 321 can be applied to the subband signal 212 (with the original resolution). As a result, DRC can be performed in an efficient manner without degrading the sound quality.

[0128] As already pointed out above, the term "compression" in this document refers to "dynamic range compression". This term is different from the term "coded", which may involve data or bitrate compression operations (for reducing the transmission rate of an audio signal).

[0129] Thus, a dynamic range control unit 210 configured to apply dynamic range control (DRC) to an audio signal 211 is described. The DRC unit 210 can include one or more units described in the context of Figure 3 . The DRC unit 210 can be part of the decoding unit 120 or can work with the decoding unit.

[0130] The DRC unit 210 can be configured to downsample the subband signal 212 obtained from the audio signal 211 to provide the downsampled subband signal 321. The downsampling can be performed using the downsampling unit 301 described in the context of Figure 3 . Generally, the audio signal 211 is represented using K subband signals 212 of K frequency bins. Downsampling can be performed on a part or all of the K subband signals 212 (especially on the K max subband signals), where K max < K. The downsampling can be performed using an (integer) downsampling factor M such that the downsampled subband signal 321 includes N / M samples compared to the corresponding subband signal 212 that includes N samples.

[0131] Specifically, a frame or block of the subband signal 212 can include N samples corresponding to N time slots, and the corresponding frame or block of the downsampled subband signal 321 can include N / M samples of N / M time slots, where M > 1 is the downsampling factor.

[0132] DRC unit 210 can be configured to downsample subband signal 212 without using a filter, particularly a low-pass filter. Specifically, DRC unit 210 can be configured to determine downsampled subband signal 321 by selecting every Mth sample of subband signal 212 as a sample of downsampled subband signal 321. By doing so, the downsampling operation can be performed particularly efficiently (this is possible because one or more downsampled subband signals 321 are used only to determine DRC gain 329).

[0133] DRC unit 210 is further configured to determine DRC gain 329 based on downsampled subband signal 321. Specifically, a sequence with N / M DRC gains 329 can be determined for N / M samples of downsampled subband signal 321. In other words, DRC unit 210 can be configured to determine N / M DRC gains 329 for the corresponding N / M time slots of downsampled subband signal 321 based on downsampled subband signal 321. The DRC gain 329 can correspond to... Figure 3 The control gain c described in the context in .

[0134] Additionally, DRC unit 210 is configured to apply DRC gain 329 to subband signal 212 to provide compressed subband signal 213 of compressed audio signal 214. For this purpose, DRC unit 210 can be configured to upsample N / M to N / M DRC gains 329 of the corresponding N samples and / or time slots of subband signal 212. Upsampling can be performed efficiently by reusing the same DRC gain 329 for M subsequent time slots. Alternatively, upsampling can be performed by interpolating between two consecutive DRC gains 329 from N / M DRC gains 329 to determine M-1 DRC gains 329 for M-1 subsequent time slots between the two consecutive DRC gains 329 from N / M DRC gains 329. In other words, an interpolation scheme can be used between consecutive (downsampled) DRC gains 329 to determine intermediate (upsampled) DRC gains 329. For example, linear interpolation or some other form of interpolation can be used, especially second-order or higher-order interpolation. However, it should be noted that upsampling using higher-order interpolation may lead to increased complexity and / or latency compared to using zero-order interpolation (e.g., repeating the same DRC gain 329 for M subsequent time slots).

[0135] After upsampling, DRC unit 210 can be configured to multiply N samples of the corresponding N time slots of subband signal 212 by the corresponding N DRC gains 329 (in the linear domain) to determine compressed subband signal 213.

[0136] Therefore, DRC can be performed in a sub-band domain in an efficient computational manner while maintaining high perceptual quality.

[0137] As described above, the audio signal 211 can be represented by K sub-band signals 212 of K different frequency bins, where K > 1 (K = 20 or more, or K = 50 or more, especially K = 64).

[0138] The DRC unit 210 can be configured to group at least some of the K different frequency bins into Q DRC bands, where Q < K, particularly Q << K (e.g., Q > 1 and / or Q < 64, e.g., Q = 4). Thus, the Q DRC bands can be formed by dispatching at least some or all of the downsampled sub-band signals 312 to different DRC bands. The DRC unit 210 can be further configured to determine the DRC gain 329 of at least one of the Q DRC bands respectively based on the sub-band signal 212 dispatched to at least one of the Q DRC bands. Particularly, dedicated DRC gains 329 can be determined for the Q different DRC bands. On the other hand, the same DRC gain can be used for the sub-band signals 212 that have been dispatched to a joint DRC band. By forming different DRC bands to determine the joint DRC gain 329 of different sub-band signals 212 of the corresponding DRC bands, the computational complexity can be further reduced.

[0139] The DRC unit 210 can be configured to ignore one or more of the K frequency bins when determining the DRC gain 329 of the Q DRC bands, especially one or more frequency bins corresponding to the highest frequency in the K frequency bins and / or corresponding to the lowest frequency in the K frequency bins (especially the lowest or DC frequency bin). In other words, when determining the DRC gain 329, some sub-band signals 212 (especially the sub-band signals 212 for frequency bins k > K max can be ignored). By doing so, the computational complexity can be further reduced.

[0140] The DRC unit 210 can be configured to determine a loudness estimate 323 of the samples of the downsampled sub-band signal 321. An example loudness estimate 323 is L inl (in the linear domain) or (in the logarithmic domain), as Figure 3 outlined in the context of. The loudness estimate 323 can be determined for each time slot or sample n of the downsampled sub-band signal 321, each DRC band i, and / or each channel l.

[0141] To determine the loudness estimate 323 of the samples of the multiple downsampled sub-band signals 321 regrouped to or dispatched to the DRC bands, the power or amplitude of the corresponding samples of the multiple downsampled sub-band signals 321 can be summed.

[0142] DRC unit 210 can be configured to determine the gain value 329 of the samples of the downsampled subband signal 321 based on the loudness estimate 323 of the samples. For this purpose, a compression curve can be used, which is configured to assign the gain value (referred to herein as the mapped gain value 324) to the loudness estimate 323 (or a value obtained from the loudness estimate 323). The compression curve may correspond to or may be the compression curve specified above (in... Figure 3 (in the context).

[0143] DRC unit 210 can be configured to determine the amplitude of a sample of the downsampled subband signal 321 (or corresponding samples of multiple downsampled subband signals 321 assigned to the same DRC frequency band). The amplitude can be in Figure 3 The value I described in the context inl .

[0144] Furthermore, DRC unit 210 can be configured to apply Stevens' power law and a transformation from the linear domain to the logarithmic domain to the amplitude of samples of the downsampled subband signal 321 to determine a loudness estimate 323 of the samples of the downsampled subband signal 321. Using Stevens' power law allows the loudness estimate 323 to be determined in a computationally efficient manner.

[0145] The compression curve can be configured to reflect the deviation between loudness estimate 323 and operating point 331 (e.g., using formula). The gain value provided is 324, which is referred to here as the mapped gain value (e.g., as shown in the image). Figure 3 The value F is summarized in the context. inl ).

[0146] The operating point 331 can depend on the dialogue normalization parameter, the average dialogue level of the dialogue content within the audio signal 211, or the average power level of the audio signal 211. This parameter can be provided in the bitstream 112 of the audio signal 211. By taking into account the operating point 331 used to apply the compression curve (which can vary along the timeline), the impact of DRC on the perceived quality of the audio signal 211 can be further reduced.

[0147] DRC unit 210 can be configured to: based on the power values ​​of blocks (or frames) of samples of the downsampled subband signal 321 (e.g., in... Figure 3 BA value as outlined in the context The average block power value is determined. Then, the operating point 331 of the sample block (or frame) of the downsampled subband signal 321 can be determined based on the average block power value.

[0148] Specifically, the DRC unit 210 can be configured to determine the long-term average block power value of the sample blocks of the downsampled subband signal 321 based on the average block power value of the sample blocks of the downsampled subband signal 321 and the sequence of previous sample blocks of the downsampled subband signal 321. Specifically, the long-term average block power value... It is possible Figure 3 As outlined in the context, the long-term average block power can be reset if a scene change is detected within the audio signal 211 (especially in the case of a sudden change in loudness estimate 323).

[0149] Furthermore, DRC unit 210 can be configured to, for example, determine the power deviation ΔL between the long-term average block power value and the white normalization parameter using the following formula. t

[0150]

[0151] The operating point 331 of the sample from the sample block of the downsampled subband signal 321 can then be accurately determined based on the power deviation, for example, using the formula As a result, the impact of DRC on the sound quality of the audio signal can be further reduced. In particular, by setting the operating point 331 for different DRC bands as outlined above, the spectral composition of the audio signal can be maintained even when DRC is performed in the DRC band (i.e., the equalization effect can be reduced).

[0152] The compression profile can be defined by profile data 215 (which can be provided by bitstream 112). Profile data 215 may include or may indicate one or more gain values ​​for implementing up compression and / or down compression. In particular, profile data 215 may include or may indicate a gain value G. maxcut G sectioncut G sectionboost and / or G maxboost Alternatively or additionally, the configuration file data 215 may include or may indicate one or more loudness thresholds. Loudness thresholds can be used to specify different loudness intervals for different segments of the compression curve loudness estimate 323. In particular, the configuration file data 215 may include or may indicate a threshold L. maxcut L sectioncut L 0high L sectionboost L 0low and / or L maxboost By providing the parameters of the compression curve as configuration file data 215, DRC can be performed efficiently and accurately (especially at decoding unit 120) without actually significantly increasing the bit rate of the codec system 100.

[0153] DRC unit 210 can be configured to: determine a mapped gain value 324 (e.g., using a compression curve) for the samples of the downsampled subband signal 321 in the current time slot based on a loudness estimate 323 of the loudness of the samples of the downsampled subband signal 321 in the current time slot. Furthermore, DRC unit 210 can be configured to: perform gain smoothing on the gain values ​​324 of the samples of the downsampled subband signal 321 in the current time slot and one or more previous time slots to determine a smoothed gain value 326 for the samples of the downsampled subband signal 321 in the current time slot. Gain smoothing can be performed as outlined in the context of gain smoothing unit 306. In particular, gain smoothing can use a low-pass filter. The smoothing factor α used to perform gain smoothing can depend on whether the loudness estimate 323 is above or below the operating point 331 of the compression curve. Alternatively or additionally, the smoothing factor can depend on whether the loudness estimate 323 at the current time slot n is above or below the smoothed gain value 326 at the previous time slot n-1 (e.g., as...). ​ (As outlined in the context). By doing so, the impact of DRC on the perceived quality of the audio signal 211 can be further reduced.

[0154] The DRC gain 329 of the samples of the downsampled subband signal 321 in the current time slot n can then be determined based on the smoothed gain value 326 of the samples of the downsampled subband signal 321 in the current time slot n. By using gain smoothing, the impact of DRC on the perceived quality of the audio signal 211 can be further reduced.

[0155] As outlined above, DRC unit 210 can be configured to determine the DRC gain 329 of a sample of the downsampled subband signal 321. The DRC gain 329 of a specific sample of the downsampled subband signal 321 can be applied to multiple consecutive samples of the subband signal 212 to determine the compressed subband signal 213. In other words, the DRC gain 329 can be upsampled to the sampling rate of the subband signal 212. Upsampling can be performed using a sample-and-hold method. As a result, DRC can be performed computationally efficiently.

[0156] Audio signal 211 may include multiple different channels. In other words, audio signal 211 may be a multi-channel audio signal. DRC unit 210 may be configured to determine the joint DRC gain 329 for multiple different channels or one or more channel subgroups. Example channel subgroups for the 5.1 multi-channel audio signal 211 may be a first subgroup of channels {L,R,C} and a second subgroup of channels {Ls,Rs}. The joint DRC gain 329 may be determined individually for each channel subgroup.

[0157] Specifically, the DRC unit 210 can be configured to select one of multiple different channels (or from a subgroup of channels) based, particularly on multiple power values ​​of samples from corresponding multiple different channels. For example, the channel requiring the highest or lowest compression level can be selected. The DRC gain 329 of the selected channel can then be determined as the joint DRC gain 329 of multiple different channels. Thus, the DRC gain 329 can be determined for a single channel, but can be applied to multiple different channels. By doing so, computational complexity can be reduced while minimizing the impact on the perceptual quality of the audio signal 211. In particular, the spatial image of the audio signal 211 can be preserved, i.e., the balance between the different channels of the audio signal 211 can be maintained.

[0158] DRC unit 210 can be configured to determine multiple DRC gains 329 and apply the multiple DRC gains 329 to corresponding multiple sub-band signals 212 to provide corresponding multiple compressed sub-band signals 329. As outlined above, a joint DRC gain 329 can be determined for one or more DRC bands. The compressed audio signal 214 can be determined by applying an inverse transform or synthesis filter bank (especially a synthesis pseudo-QMF bank) to the multiple compressed sub-band signals 329. The inverse transform or synthesis filter bank is the complementary inverse operation of the forward transform or analysis filter bank used to obtain multiple sub-band signals 212 from the audio signal 211. In particular, the forward transform and inverse transform pair and / or a pair of analysis filter banks and synthesis filter banks can meet the perfect reconstruction criterion.

[0159] DRC unit 210 can be configured to determine the DRC gain 329 of samples of (downsampled) subband signals 212, 321 based on a reference level of the compressed subband signal 329. The reference level can be indicated in bitstream 112 (e.g., as profile data 215). The reference level can be used to cancel out (potentially smoothed) gains 324, 326 provided by the compression curve, in order to adjust the compressed audio signal 214 according to the reference level. Specifically, an offset B 332 (such as...) can be applied. ​ (As outlined in the context). As a result, the expected loudness level of the audio signal can be set to the desired value.

[0160] Furthermore, a decoding unit 120 is described, configured to obtain a dynamic range compressed audio signal 214 from a bitstream 112. The bitstream 112 may indicate an audio signal 211 or multiple sub-band signals 212 of the audio signal 211. The decoding unit 120 includes a DRC unit 210 as described in this document. The DRC unit 210 is configured to apply dynamic range compression to the audio signal 211.

[0161] This document describes a scheme that allows DRC to be performed in a computationally efficient manner without degrading sound quality.

[0162] The methods and systems described in this document can be implemented as software, firmware, and / or hardware. Certain components can be implemented, for example, as software running on a digital signal processor or microprocessor. Other components can be implemented, for example, as hardware and / or application-specific integrated circuits (ASICs). Signals encountered in the described methods and systems can be stored on media such as random access memory or optical storage media. These signals can be transmitted via networks such as radio networks, satellite networks, wireless networks, or wired networks (e.g., the Internet). Typical devices utilizing the methods and systems described in this document are portable electronic devices or other consumer devices for storing and / or presenting audio signals.

Claims

1. A dynamic range control (DRC) unit (210) configured to apply DRC to an audio signal (211); wherein, The DRC unit (210) is configured as follows: - The sub-band signal (212) obtained from the audio signal (211) is downsampled to provide a downsampled sub-band signal (321). - Determine the DRC gain (329) based on the downsampled subband signal (321); as well as - Apply the DRC gain (329) to the subband signal (212) to provide a compressed subband signal (213) of the compressed audio signal (214).

2. The DRC unit (210) as described in claim 1, wherein, -The frames of the subband signal (212) include N Each time slot, and the corresponding frame of the downsampled subband signal (321) includes One time slot; and - M > 1 is the downsampling factor.

3. The DRC unit (210) as described in claim 2, wherein, The DRC unit (210) is configured as follows: -Based on the downsampled sub-band signal (321), determine the corresponding... Each time slot DRC gain (329); as well as -Regarding the above Each DRC gain (329) is applied to the corresponding subband signal (212). N Each time slot N Upsampling of DRC gain (329).

4. The DRC unit (210) as described in claim 3, wherein, Upsampling includes -right M The same DRC gain (329) is reused in subsequent time slots; or -in the source Interpolation is performed between two consecutive DRC gains (329) of the DRC gain (329) to determine the value from the DRC gain (329). Between two consecutive DRC gains (329) of DRC gain (329) M - 1 subsequent time slot M - 1 DRC gain (329).

5. The DRC unit (210) as described in claim 3 or 4, wherein, The DRC unit (210) is configured to: convert the corresponding subband signal (212) N Each time slot N Each sample and its corresponding N The DRC gain (329) is multiplied to determine the compressed subband signal (213).

6. The DRC unit (210) as described in claim 1 or 2, wherein, The DRC unit (210) is configured as follows: - Downsample the subband signal (212) without using a low-pass filter; and / or - By selecting each of the subband signals (212) M One sample is used as a sample of the downsampled subband signal (321) to determine the downsampled subband signal (321).

7. The DRC unit (210) as described in claim 1 or 2, wherein, -The audio signal (211) uses K Different frequency chambers K The sub-band signal (212) is used to represent this, where, K > 1; and - The DRC unit (210) is configured as follows: -to the K At least some of the groups in the different frequency bins are Q One DRC frequency band, of which Q < K ;as well as -Based on the assignment to the Q The sub-band signal (212) of at least one DRC frequency band in the DRC frequency band determines the... Q DRC gain (329) of at least one DRC band in the DRC bands.

8. The DRC unit (210) as claimed in claim 7, wherein, The DRC unit (210) is configured to determine the Q When the DRC gain (329) of each DRC band is ignored, the above K One or more of different frequency bins.

9. The DRC unit (210) as claimed in claim 8, wherein, The one or more ignored frequency bins correspond to the K The highest frequency in each of the different frequency chambers.

10. The DRC unit (210) as claimed in claim 1 or 2, wherein, The DRC unit (210) is configured as follows: - Determine the loudness estimate (323) of the sample of the downsampled subband signal (321); and - The DRC gain (329) is determined based on the loudness estimate (323) using the compression curve.

11. The DRC unit (210) as claimed in claim 10, wherein, The DRC unit (210) is configured as follows: - Determine the amplitude of the sample of the downsampled subband signal (321); and - Apply Stevens' power law and the transformation from the linear domain to the logarithmic domain to the amplitude of the samples of the downsampled subband signal (321) to determine the loudness estimate (323) of the samples of the downsampled subband signal (321).

12. The DRC unit (210) as claimed in claim 10, wherein, - The compression curve is configured to provide a mapping gain value (324) based on the deviation between the loudness estimate (323) and the operating point (331); and - The operating point (331) depends on the dialogue normalization parameter or the average dialogue level of the dialogue content in the audio signal (211) or the average power level of the audio signal (211).

13. The DRC unit (210) as claimed in claim 12, wherein, The DRC unit (210) is configured as follows: - Determine the average block power value based on the power value of the sample block of the downsampled subband signal (321); and -Based on the average block power value, the operating point (331) of the sample block of the downsampled subband signal (321) is determined.

14. The DRC unit (210) as claimed in claim 13, wherein, The DRC unit (210) is configured as follows: -Based on the average block power value of the sample blocks of the downsampled subband signal (321) and the sequence of previous sample blocks of the downsampled subband signal (321), determine the long-term average block power value of the sample blocks of the downsampled subband signal (321). - Determine the power deviation between the long-term average block power value and the normalized parameter; and -Based on the power deviation, determine the operating point (331) of the sample in the sample block from the downsampled subband signal (321).

15. The DRC unit (210) as claimed in claim 10, wherein, The compression curve is defined by one or more parameters included in the configuration file data (215).

16. The DRC unit (210) as claimed in claim 1 or 2, wherein, The DRC unit (210) is configured as follows: - Based on the loudness estimation (323) of the samples of the downsampled subband signal (321) of the current time slot, determine the mapping gain value (324) of the samples of the downsampled subband signal (321) of the current time slot. - Perform gain smoothing on the mapped gain values ​​(324) of the samples of the downsampled subband signal (321) in the current time slot and one or more previous time slots to determine the smoothed gain value (326) of the samples of the downsampled subband signal (321) in the current time slot. as well as -Based on the smoothing gain value (326) of the sample of the downsampled subband signal (321) in the current time slot, determine the DRC gain (329) of the sample of the downsampled subband signal (321) in the current time slot.

17. The DRC unit (210) as claimed in claim 1 or 2, wherein, The DRC unit (210) is configured as follows: - Determine the DRC gain (329) of the samples of the downsampled subband signal (321); and - The DRC gain (329) of the samples of the downsampled subband signal (321) is applied to a plurality of consecutive samples of the subband signal (212) to determine the compressed subband signal (213).

18. The DRC unit (210) as claimed in claim 1 or 2, wherein, - The audio signal (211) includes multiple different channels; and - The DRC unit (210) is configured to: determine the combined DRC gain (329) of the plurality of different channels or one or more subgroups of the plurality of different channels.

19. The DRC unit (210) as claimed in claim 18, wherein, The DRC unit (210) is configured as follows: - Select one of the multiple different channels; as well as - Determine the DRC gain (329) of the selected channel as the combined DRC gain (329) of the plurality of different channels or one or more subgroups of the plurality of different channels.

20. The DRC unit (210) as claimed in claim 19, wherein, The DRC unit (210) is configured to select one of the multiple different channels based on multiple power values ​​of samples corresponding to the multiple different channels.

21. The DRC unit (210) as claimed in claim 1 or 2, wherein, The DRC unit (210) is configured as follows: - Determine and apply multiple DRC gains (329) for the corresponding multiple subband signals (212) to provide the corresponding multiple compressed subband signals (213); and -The compressed audio signal (214) is determined by applying an inverse transform or synthesis filter bank to the plurality of compressed subband signals (213); The inverse transform or the synthesis filter bank is complementary to the forward transform or the analysis filter bank, and the forward transform or the analysis filter bank is used to obtain the sub-band signal (212) from the audio signal (211).

22. A decoding unit (120) configured to: obtain a dynamic range compressed audio signal (214) from a bitstream (112) indicating an audio signal (211); wherein, The decoding unit (120) includes a DRC unit (210) according to any one of the preceding claims, the DRC unit being configured to apply dynamic range control to the audio signal (211).

23. A method (400) for applying dynamic range control (DRC) to an audio signal (211); wherein, The method (400) includes - The sub-band signal (212) obtained from the audio signal (211) is downsampled (401) to provide a downsampled sub-band signal (321). - Determine (402) DRC gain (329) based on the downsampled subband signal (321); as well as - Apply the DRC gain (329) (403) to the subband signal (212) to provide a compressed subband signal (213) of the compressed audio signal (214).

24. A storage medium comprising a software program that is executed on a processor and, when executed by the processor, is used to perform the method of claim 23.

25. A computer program product comprising executable instructions that, when executed on a computer, are configured to perform the method of claim 23.

Images