Efficient Sample Rate Conversion

Abstract

A method (500) for resampling an audio signal (110) is described. The method (500) comprising providing (501) a set of input subband signals (210) which is representative of a time domain audio signal. Furthermore, the method (500) comprises applying (502) a first ripple pre-emphasis gain (323) to a first input subband signal (210) of the set of input subband signals (210) to determine a corresponding first output subband signal (213) of a set of output subband signals (213). In addition, the method (500) comprises determining (503) a time domain input audio signal (110) from the set of output subband signals (213). The method (500) further comprises performing (504) time domain resampling of the input audio signal (110) to provide an output audio signal (113) using an anti-aliasing filter (102), wherein the first ripple pre-emphasis gain (323) is dependent on a frequency response (311) of the anti-aliasing filter (102), such that an amplitude ripple of the frequency response (311) of the anti-aliasing filter (102) is at least partially compensated by the first ripple pre-emphasis gain (323).

Claims

1. A method for resampling an audio signal, the method comprising providing a set of input subband signals which is representative of a time domain audio signal; applying a first ripple pre-emphasis gain to a first input subband signal of the set of input subband signals to determine a corresponding first output subband signal of a set of output subband signals; determining a time domain input audio signal from the set of output subband signals; and performing time domain resampling of the input audio signal to provide an output audio signal using an anti-aliasing filter; wherein the first ripple pre-emphasis gain is dependent on a frequency response of the anti-aliasing filter, such that an amplitude ripple of the frequency response of the anti-aliasing filter is at least partially compensated by the first ripple pre-emphasis gain.

2. The method of claim 1, wherein the first input subband signal corresponds to a first subband of a plurality of subbands covering different frequency ranges; and the first ripple pre-emphasis gain is dependent on the frequency response of the anti-aliasing filter within the first subband.

3. The method of claim 2, wherein the first ripple pre-emphasis gain is dependent on an average amplitude or a median amplitude of the frequency response of the anti-aliasing filter within the first subband.

4. The method of claim 2, wherein the first ripple pre-emphasis gain is such that it at least partially compensates a deviation of an amplitude of the frequency response of the anti-aliasing filter within the first subband from a reference value.

5. The method of claim 1, wherein the method comprises applying different ripple pre-emphasis gains to the different input subband signals; and the different ripple pre-emphasis gains are such that the ripple pre-emphasis gains at least partially compensate the amplitude ripple of the frequency response of the anti-aliasing filter.

6. The method of claim 1, wherein the input subband signals of the set of input subband signals correspond to different subbands of a plurality of subbands; the method comprises applying different ripple pre-emphasis gains to the different input subband signals; the ripple pre-emphasis gains for the different subbands form a frequency-dependent gain curve; the frequency response of the anti-aliasing filter exhibits a frequency-dependent amplitude curve; the different ripple pre-emphasis gains are such that the frequency-dependent gain curve approximates an inverse of the frequency-dependent amplitude curve; and/or a deviation of the frequency-dependent gain curve from the inverse of the frequency-dependent amplitude curve is equal to or smaller than a deviation threshold; and/or a product of the frequency-dependent gain curve and the frequency-dependent amplitude curve provides a frequency-dependent product curve having a variance which is lower than a variance of the frequency-dependent amplitude curve.

7. The method of claim 1, wherein the method comprises determining the first ripple pre-emphasis gain based on the frequency response of the anti-aliasing filter.

8. The method of claim 1, wherein the time domain input audio signal is determined using a synthesis filterbank comprising Q synthesis filters; and the first ripple pre-emphasis gain is dependent on a power response of at least one of the Q synthesis filters.

9. The method of claim 8, wherein G is a row vector comprising L squared ripple pre-emphasis gains for L input subband signals; P is a matrix with size LxK, which is indicative of a power response of L analysis-synthesis filter combinations at K pass band frequency points; and T is a row vector indicative of a power response of the anti-aliasing filter at the K pass band frequency points; and G is determined such that G×P=1/T is approximated in a least squares error sense.

10. The method of claim 1, wherein the set of input subband signals is provided from encoded audio data of a bitstream.

11. The method of claim 1, wherein providing the set of input subband signals comprises applying an analysis filterbank comprising a plurality of analysis filters to a time domain audio signal, and optionally, wherein the analysis filters comprise quadrature mirror filters; and/or a number of analysis filters is 32, 64 or more.

12. The method of claim 11, wherein the time domain audio signal is a lowband audio signal; the method comprises determining a set of lowband subband signals by applying the analysis filterbank to the lowband audio signal; and the method comprises performing spectral band expansion using the set of lowband subband signals and one or more spectral band expansion parameters to provide the set of input subband signals.

13. The method of claim 1, wherein determining the time domain input audio signal comprises applying a synthesis filterbank comprising a plurality of synthesis filters to a set of subband signals derived from the set of output subband signals.

14. The method of claim 13, wherein the synthesis filters comprise quadrature mirror filters; and/or a number of analysis filters is 32, 64 or more.

15. The method of claim 13, wherein the method comprises upmixing the set of output subband signals to a first and a second set of output subband signals using one or more upmixing parameters; and deriving the set of subband signals to which the synthesis filterbank is applied from the first set of output subband signals.

16. A system for determining a resampled output audio signal, wherein the system comprises, a transform unit configured to provide a set of input subband signals which is representative of a time domain audio signal; a pre-emphasis unit configured to apply a first ripple pre-emphasis gain to a first input subband signal of the set of input subband signals to determine a corresponding first output subband signal of a set of output subband signals; an inverse transform unit configured to determine a time domain input audio signal from the set of output subband signal; and a resampling unit configured to perform time domain resampling of the input audio signal to provide an output audio signal using an anti-aliasing filter; wherein the first ripple pre-emphasis gain is dependent on a frequency response of the anti-aliasing filter, such that an amplitude ripple of the frequency response of the anti-aliasing filter is at least partially compensated by the first ripple pre-emphasis gain.

17. The system of claim 16, wherein the transform unit and the inverse transform unit form a perfect reconstruction filterbank or a near perfect reconstruction filterbank.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0046] The methods and systems described in the present document are explained below in an exemplary manner with reference to the accompanying drawings, wherein

[0047] FIG. 1 illustrates a conceptual diagram of an example time domain resampler;

[0048] FIG. 2 shows a block diagram of an example audio decoder comprising a sample rate converter or time domain resampler;

[0049] FIG. 3a shows example magnitude frequency responses of different anti-aliasing filters of a time domain resampler;

[0050] FIG. 3b shows example magnitude frequency responses of different anti-aliasing filters within a pass band;

[0051] FIG. 4 shows the effective frequency response of an anti-aliasing filter when using subband domain ripple pre-emphasis; and

[0052] FIG. 5 shows a flow chart of an example method for resampling an audio signal.

DETAILED DESCRIPTION OF THE FIGURES

[0053] FIG. 1 illustrates conceptually a time domain resampler 100 for rational resampling factors N/M. The time domain resampler 100 comprises an up-by-N upsampler 101 which converts a sample of an input audio signal 110 into N samples of an upsampled audio signal 111. This may be achieved by inserting N-1 zeros between two samples of the input audio signal 110. Subsequently, the upsampled audio signal is filtered by an anti-aliasing/anti-imaging filter 102 with transfer function H(z). This results in a filtered audio signal 112. Finally, the filtered audio signal 112 is passed to a down-by-M decimator or downsampler 103 which only retains every M.sup.th sample of the filtered audio signal 112, to thereby provide the resampled (or output) audio signal 113. In case of a resampling of an input audio signal 110 at a sampling rate of 32 kHz to an output audio signal 113 at a sampling rate of 48 kHz, N is 3 and M is 2. In case of a resampling of an input audio signal 110 at a sampling rate of 44.1 kHz to an output audio signal 113 at a sampling rate of 48 kHz, N is 160 while M is 147.

[0054] It should be noted that the filter 102 runs at an intermediate frequency (IF) at N times the input sampling rate or at M times the output sampling rate (e.g. IF=M*48 kHz for the above mentioned cases). This means that the anti-aliasing filters 102 typically operate at high sampling rates, such that a reduction of the number of computational filter operations is desirable. In other words, it is desirable to reduce the number of required coefficients of the anti-aliasing filter 102, in order to reduce the overall computational complexity of the time domain resampler 100.

[0055] The filters 102 may be realized as a polyphase FIR (Finite Impulse Response) implementation. Such an implementation exploits the fact that the upsampled audio signal 111 which is filtered by filter 102 comprises N−1 zeros between the samples of the input audio signal 110. Consequently, the “zero” multiplications and additions can be omitted. Furthermore, a polyphase implementation exploits the fact that due to the subsequent down-by-M decimator 103, only every M.sup.th sample of the filtered audio signal 112 needs to be determined. By exploiting this information during the filter implementation, the number of multiplication and/or adding operations can be significantly reduced, thereby reducing the computational complexity of the time domain resampler 100. Nevertheless, it is desirable to further reduce the computational complexity or to further improve the perceptual performance of the resampler 100.

[0056] As indicated above, the resampling operation creates imaging and/or aliasing artifacts in the output audio signal 113 if no anti-aliasing filter 102 is used. These imaging and/or aliasing artifacts are created as a result of the upsampling 101 and downsampling 103 operations. An anti-aliasing filter 102 may be designed by defining a target frequency response of the anti-aliasing filter 102 such that imaging and/or aliasing artifacts in the output audio signal 113 are avoided. The filter coefficients of a filter 102 meeting or approximating such target frequency response may be determined using filter design methods such as the Parks-McClellan algorithm. This algorithm determines the set of filter coefficients that minimize the maximum deviation from the target frequency response.

[0057] The Parks-McClellan algorithm is directed at minimizing the maximum of an approximation error E(f) given by

E(f)=W(f)|D(f)−H(f)|,

wherein D(f) is the desired form of the low pass filter 102, i.e. the target frequency response, and is typically given by

[00001]$D\left(f\right)=\{\begin{array}{c}1,f\in \left[0,f_p\right]\\ 0,f\in \left[f_s,{\mathrm{Mfs}}_{\mathrm{out}}/2\right]\end{array},$

with f.sub.p being the pass band edge and f.sub.s being the stop band edge. W(f) is a frequency dependent weighting function of the approximation error. In particular W(f) may comprise one or more pass band weights for the pass band and one or more stop band weights for the stop band of the filter 102. H(f) is given by

[00002]$H\left(f\right)=\underset{k=0}{\overset{n}{.Math.}}.Math.d_k.Math.\cos \left(2.Math.k.Math..Math.\pi .Math..Math.f\right),$

and relates to the frequency response of the filter 102 by exp(−j2πnf)H(f). The filter coefficients h.sub.k of filter 102 are given by

h.sub.k=h.sub.2n-k; d.sub.n-k=2h.sub.k, k=0, . . . , n−1; d.sub.0=h.sub.n.

[0058] Details on the Parks-McClellan algorithm are outlined in T. Parks, J. McClellan, “Chebyshev Approximation for Nonrecursive Digital Filters with Linear Phase”, IEEE Transactions on Circuit Theory, Vol. CT-19, No. 2, March 1972, which is incorporated by reference.

[0059] Typically, the quality of an anti-aliasing filter 102 (notably with respect to an amplitude ripple within the pass band) increases with an increasing number of filter coefficients (also referred to as filter taps). FIG. 3a shows the frequency response, i.e. the gain or attenuation 302 of the filter 102 as a function of frequency 301, for different anti-aliasing filters 102 having different numbers of filter taps. In particular, FIG. 3a shows the frequency response 311 of a filter 102 having 12 taps, the frequency response 312 of a filter having 16 taps and the frequency response 313 of a filter having 20 taps. FIG. 3b shows the amplitude ripple of the frequency responses 311, 312, 313 within the pass band. It can be seen that with increasing number of taps, the amplitude ripple of the frequency responses 311, 312, 313 (i.e. the deviation of the gain/attenuation 302 of the frequency responses 311, 312, 313 from a reference value of e.g. 1 or 0 dB) may be decreased. On the other hand, the computational complexity increases with an increasing number of filter taps.

[0060] In various different applications, an audio signal 110, which is to be re-sampled, may be submitted to subband processing, which involves processing of the audio signal 110 in a subband domain (e.g. in a QMF, Quadrature Mirror Filter, domain). An example application is an audio decoder such as an AC-4 audio decoder which makes use of subband processing for expanding the bandwidth of an audio signal. FIG. 2 shows a block diagram of such an example audio decoder 200. The audio decoder 200 comprises a bitstream demultiplexing unit 201 which is configured to provide audio data 221 for a lowband audio signal 223 from a bitstream received at the decoder 200. Furthermore, the bitstream demultiplexing unit 201 is configured to provide metadata 222 which enables e.g. decoupling and/or upmixing of an audio signal (to generate a plurality of audio channels from a downmixed audio channel) and/or spectral band expansion (to expand the bandwidth of the lowband audio signal).

[0061] The audio data 221 may be provided to an audio decoding unit 202 which is configured to generate the lowband audio signal 223 from the audio data 221. The lowband audio signal 223 it typically a time domain audio signal (e.g. PCM encoded). The audio decoder 200 comprises a transform unit 203 which is configured to convert the time-domain lowband audio signal 223 into a subband domain signal comprising a set of lowband subband signals 224. The set of lowband subband signals 224 may be submitted to spectral band expansion using the metadata 222 within a spectral band expansion unit 204. In particular, a set of highband subband signals may be generated from the set of lowband subband signals 224 to provide an overall set of subband signals 210 (also referred to herein as a set of input subband signals 210).

[0062] The set of subband signals 210 or a modified version 213 thereof may be submitted to decoupling within a decoupling unit 206 to provide multiple sets of subband signals 225 from a single set of subband signals 210, 213 for multiple audio channels, respectively. Furthermore, the one or more sets of subband signals 225 may be submitted to optional subband procession within a subband processing unit 207. Furthermore, the one or more sets of subband signals may be converted from the subband domain into the time domain using an inverse transform unit 208, thereby providing one or more audio signals 110 for one or more channels.

[0063] In the following an example audio signal 110 provided by the decoder 200 is considered. This time domain audio signal 110 is obtained from a respective set of subband signals 210 using an inverse transform or a synthesis filterbank which is applied in the inverse transform unit 208. By way of example, the transform unit 203 and the inverse transform unit 208 comprise a QMF filter bank (e.g. with Q=32 or 64 subbands).

[0064] There may be a need for resampling the audio signal 110 (also referred to as the input audio signal 110) using a rational resampler 100, thereby providing an output audio signal 113 at a different sampling rate. As outlined above, the resampler 100 typically comprises an anti-aliasing filter 102 having a certain frequency response 311, 312, 313 and exhibiting a varying gain/attenuation 302 within the pass band of the filter 102. In an ideal case, the anti-aliasing filter 102 exhibits a constant gain at a reference value (e.g. 0 dB or 1) across the entire pass band (i.e. up to the pass band edge f.sub.p). However, actual anti-aliasing filters 102 exhibit a deviation from this constant gain, as can be seen in FIG. 3b. This deviation may be referred to as ripple or amplitude ripple.

[0065] The decoder 200 may comprise a ripple pre-emphasis unit 205 which is configured to apply ripple pre-emphasis gains to the different subband signals of a set of input subband signals 210. In particular, a subband-dependent ripple pre-emphasis gain may be applied to some or all of the input subband signals 210, thereby providing a set of output subband signals 213. The set of output subband signals 213 (or a processed version thereof) may then be submitted to the inverse transform unit 208 to derive the audio signal 110 which is to be resampled within the resampler 100.

[0066] The ripple pre-emphasis gains may determined based on the frequency response 311 of the anti-aliasing filter 102 used within the resampler 100. FIG. 3b illustrates the frequency range of different subbands 321 of the subband domain along the frequency axis. It can be seen that different subbands 321 correspond to different spectral excerpts of the frequency response 311 of the anti-aliasing filter 102. Furthermore, it can be seen that the frequency response 311 of the anti-aliasing filter 102 exhibits a different gain/attenuation 302 within the different subbands 321. For some or all of the subbands 321 an effective gain 322 of the frequency response 311 of the anti-aliasing filter 102 (e.g. as the mean value of the gain/attenuation 302 within the different subbands 321) may be determined. The ripple pre-emphasis gain 323 for a particular subband 321 may correspond to the complementary value of the effective gain 322 of the frequency response 311 of the anti-aliasing filter 102 for the particular subband 321. In particular, the ripple pre-emphasis gain 323 a.sub.r in dB may be a.sub.r=−a.sub.e, with a.sub.e being the effective gain 322 in dB.

[0067] As such, the ripple which is caused by the anti-aliasing filter 102 of the resampler 100 may be compensated by applying a ripple pre-emphasis gain 323 to the subband signals 210 in the subband domain, thereby providing an input audio signal 110 to the resampler 100 which already takes into account the ripple which is caused by the anti-aliasing filter 102 of the resampler 100. As a result of this, the ripple of the anti-aliasing filter 102 does not affect the output audio signal 113. In other words, an artifact-free output audio signal 113 may be generated by performing ripple pre-emphasis within the subband domain.

[0068] The use of ripple pre-emphasis within the subband domain enables the use of anti-aliasing filters 102 with a reduced number of filter taps. FIG. 4 shows the frequency response 311 of an anti-aliasing filter 102 having 12 taps and the frequency response 313 of an anti-aliasing filter 102 having 20 taps (without using ripple pre-emphasis within the subband domain). Furthermore, FIG. 4 shows the effective frequency response 411 of an anti-aliasing filter 102 having 12 taps in combination with ripple pre-emphasis. It can be seen that the frequency response 411 is almost constant within the pass band, even though an anti-aliasing filter 102 having only 12 taps has been used within the resampler 100. Hence, ripple pre-emphasis may be used to reduce the computational complexity of resampling, while maintaining a high perceptual quality for resampling, since applying pre-emphasis gains in the subband domain is computationally cheaper than adding filter taps to reduce the pass band ripple in the time domain filter.

[0069] In particular, resampling at a reduced computational complexity and good quality (in terms of low amplitude ripple and low aliasing energy) may be achieved within an audio decoder which employs QMF filter bank analysis-synthesis processing (as is the case e.g. for an AC4 codec). Amplitude ripple pre-emphasis may be applied within the QMF domain followed by time domain resampling subsequent to the QMF synthesis (within the inverse transform unit 208). Since the overall amplitude ripple is reduced by the QMF domain pre-emphasis (within the ripple pre-emphasis unit 205), the one or more time domain resampling filters 102 may be designed with reduced constraints regarding maximum amplitude ripple. Hence, for a given filter order the stop band attenuation may be increased (thereby achieving an increased quality at a constant complexity) or for a desired stop band attenuation the filter order may be reduced while maintaining low overall amplitude ripple.

[0070] The proposed scheme exploits the fact that amplitude modifications at high quality and low complexity may be achieved by the application of a subband-dependent gain within the QMF domain. Since typically the QMF domain is oversampled by a factor of 2 (due to the use of complex valued subband signals), the application of the ripple pre-emphasis gain 323 may require two multiplications per QMF subband. The QMF band ripple pre-emphasis gain 323 may be calculated as the inverse of the time domain filter magnitude response 302 within the pass band or it can be calculated in a least squares sense taking into account the filter bank net effect within a subband 321.

[0071] The anti-aliasing filter 102 may be designed such that the ripple amplitude of the filter 102 exhibits a period across frequency 301 which is substantially larger than the bandwidth of the subbands 321 of the subband transform (e.g. the QMF filterbank), e.g. 10, 15, 20 or more times larger. The bandwidth of the subbands 321 corresponds to the sample rate fs.sub.in of the input audio signal 110 divided by two times the number Q of subbands of the subband transform. As a result of such a filter design, ripple pre-emphasis may be applied in a reliable and precise manner.

[0072] By way of example, using the Parks-McClellan algorithm, the ripple period in the pass band may be increased by increasing the weight W(f) of the design accuracy of the stop band higher relative to the design accuracy of the pass band design. An iterative search may be used to determine the relative weight W(f) which provides the desired stop band attenuation and a reasonable pass band ripple which can be well compensated by QMF band ripple pre-emphasis gains.

[0073] Hence, complexity savings may be achieved by performing amplitude ripple pre-emphasis in the QMF domain, thereby reducing the requirements regarding amplitude ripple on the time domain anti-aliasing filter 102 of the resampler 100. In case of upsampling, the complexity is reduced further, due to the fact that ripple pre-emphasis is applied to the signal having the low sampling rate fs.sub.in. In case of downsampling or in case of limited signal bandwidth only a fraction of the subbands 321 need to be processed using the ripple pre-emphasis gains 323. Furthermore, in case of parametric upmixing (within the decoupling unit 206) processing complexity can be reduced, because the ripple pre-emphasis gains 323 may be applied to the downmixed channels (wherein the anti-aliasing filter 102 is applied to the upmixed channels).

[0074] In an example use case, resampling by 16/15 is performed using a 64 band QMF filterbank. The input audio signal 110 is band limited to 50/64, with 64 corresponding to the Nyquist frequency fs.sub.in/2. The conventional resampler 100 using a filter 102 with 16 filter taps requires 2×16 Multiply&Add-operations per output sample plus linear interpolation. The proposed scheme makes use of a filter 102 with 12 filter taps, which requires 2×12 Multiply&Add-operations per output sample plus linear interpolation. Furthermore, the ripple pre-emphasis requires 2×50/64×15/16=1.5 Multiply-operations per output sample. Both methods exhibit roughly the same stop band attenuation and overall peak-to-peak amplitude ripple. However, the proposed scheme enables a complexity reduction of roughly 6 Multiply&Add-operations per output sample.

[0075] FIG. 5 shows a flow chart of an example method 500 for resampling an audio signal 110. The method 500 comprising providing 501 a set of input subband signals 210. The set of input subband signals 210 may be provided by applying the analysis filters of an analysis filterbank to a time domain audio signal (e.g. to a lowband audio signal decoded using audio data 221 from a bitstream).

[0076] Furthermore, the method 500 comprises applying 502 a first ripple pre-emphasis gain 323 to a first input subband signal 210 of the set of input subband signals 210 to determine a corresponding first output subband signal 213 of a set of output subband signals 213. Typically, different ripple pre-emphasis gains 323 are applied to different input subband signals 210 of the set of input subband signals 210, thereby providing corresponding different output subband signals 210. The different ripple pre-emphasis gains 323 are frequency or subband dependent.

[0077] In addition, the method 500 comprises determining 503 a time domain input audio signal 110 from the set of output subband signals 213. This may be achieved by applying the synthesis filters of a synthesis filterbank to the set of output subband signals 213 or to a set of subband signals 225 derived from the set of output subband signals 213. The method 500 further comprises performing 504 time domain resampling of the input audio signal 110 to provide an output audio signal 113 using an anti-aliasing filter 102. The first ripple pre-emphasis gain 323 is typically dependent on and/or determined based on the frequency response 311 of the anti-aliasing filter 102. By making use of ripple pre-emphasis gains 323 which are applied in the subband domain, relatively short anti-aliasing filters 102 may be used, without increasing the effects of amplitude ripple on the output audio signal 113, thereby reducing the computational complexity for resampling.

[0078] The methods and systems described in the present document may be implemented as software, firmware and/or hardware. Certain components may e.g. be implemented as software running on a digital signal processor or microprocessor. Other components may e.g. be implemented as hardware and or as application specific integrated circuits. The signals encountered in the described methods and systems may be stored on media such as random access memory or optical storage media. They may be transferred via networks, such as radio networks, satellite networks, wireless networks or wireline networks, e.g. the internet. Typical devices making use of the methods and systems described in the present document are portable electronic devices or other consumer equipment which are used to store and/or render audio signals. The methods and system may also be used on computer systems, e.g. internet web servers, which store and provide audio signals, e.g. music signals, for download.

[0079] Various aspects of the present invention may be appreciated from the following enumerated example embodiments (EEEs): [0080] EEE 1. A method (500) for resampling an audio signal (110), the method (500) comprising [0081] providing (501) a set of input subband signals (210); [0082] applying (502) a first ripple pre-emphasis gain (323) to a first input subband signal (210) of the set of input subband signals (210) to determine a corresponding first output subband signal (213) of a set of output subband signals (213); [0083] determining (503) a time domain input audio signal (110) from the set of output subband signals (213); and [0084] performing (504) time domain resampling of the input audio signal (110) to provide an output audio signal (113) using an anti-aliasing filter (102); wherein the first ripple pre-emphasis gain (323) is dependent on a frequency response (311) of the anti-aliasing filter (102). [0085] EEE 2. The method (500) of EEE 1, wherein [0086] the first input subband signal (210) corresponds to a first subband of a plurality of subbands covering different frequency ranges; and [0087] the first ripple pre-emphasis gain (323) is dependent on the frequency response (311) of the anti-aliasing filter (102) within the first subband. [0088] EEE 3. The method (500) of EEE 2, wherein the first ripple pre-emphasis gain (323) is dependent on an average amplitude or a median amplitude of the frequency response (311) of the anti-aliasing filter (102) within the first subband. [0089] EEE 4. The method (500) of any of EEEs 2 to 3, wherein the first ripple pre-emphasis gain (323) is such that it at least partially compensates a deviation of an amplitude of the frequency response (311) of the anti-aliasing filter (102) within the first subband from a reference value. [0090] EEE 5. The method (500) of any previous EEE, wherein [0091] the method comprises applying different ripple pre-emphasis gains (323) to the different input subband signals (210); and [0092] the different ripple pre-emphasis gains (323) are such that the ripple pre-emphasis gains (323) at least partially compensate an amplitude ripple of the frequency response (311) of the anti-aliasing filter (102). [0093] EEE 6. The method (500) of any previous EEEs, wherein [0094] the input subband signals (210) of the set of input subband signals (210) correspond to different subbands of a plurality of subbands; [0095] the method (500) comprises applying different ripple pre-emphasis gains (323) to the different input subband signals (210); [0096] the ripple pre-emphasis gains (323) for the different subbands form a frequency-dependent gain curve; [0097] the frequency response (311) of the anti-aliasing filter (102) exhibits a frequency-dependent amplitude curve; [0098] the different ripple pre-emphasis gains (323) are such that [0099] the frequency-dependent gain curve approximates an inverse of the frequency-dependent amplitude curve; and/or [0100] a deviation of the frequency-dependent gain curve from the inverse of the frequency-dependent amplitude curve is equal to or smaller than a deviation threshold; and/or [0101] a product of the frequency-dependent gain curve and the frequency-dependent amplitude curve provides a frequency-dependent product curve having a variance which is lower than a variance of the frequency-dependent amplitude curve. [0102] EEE 7. The method (500) of any previous EEEs, wherein the method (500) comprises determining the first ripple pre-emphasis gain (323) based on the frequency response (311) of the anti-aliasing filter (102). [0103] EEE 8. The method (500) of any previous EEEs, wherein [0104] the time domain input audio signal (110) is determined using a synthesis filterbank comprising Q synthesis filters; and [0105] the first ripple pre-emphasis gain (323) is dependent on a power response of at least one of the Q synthesis filters. [0106] EEE 9. The method (500) of EEE 8, wherein [0107] G is a row vector comprising L squared ripple pre-emphasis gains (323) for L input subband signals (210); [0108] P is a matrix with size L×K, which is indicative of a power response of L analysis-synthesis filter combinations at K pass band frequency points; and [0109] T is a row vector indicative of a power response of the anti-aliasing filter (102) at the K pass band frequency points; and [0110] G is determined such that G×P=1/T is approximated in a least squares error sense. [0111] EEE 10. The method (500) of any previous EEEs, wherein the set of input subband signals (210) is provided from encoded audio data (221) of a bitstream. [0112] EEE 11. The method (500) of any previous EEEs, wherein providing (501) the set of input subband signals (210) comprises applying an analysis filterbank comprising a plurality of analysis filters to a time domain audio signal. [0113] EEE 12. The method (500) of EEE 11, wherein [0114] the analysis filters comprise quadrature mirror filters; and/or [0115] a number of analysis filters is 32, 64 or more. [0116] EEE 13. The method (500) of any of EEEs 11 to 12, wherein [0117] the time domain audio signal is a lowband audio signal; [0118] the method (500) comprises determining a set of lowband subband signals (224) by applying the analysis filterbank to the lowband audio signal; and [0119] the method (500) comprises performing spectral band expansion using the set of lowband subband signals (224) and one or more spectral band expansion parameters to provide the set of input subband signals (210). [0120] EEE 14. The method (500) of any previous EEEs, wherein determining (503) the time domain input audio signal (110) comprises applying a synthesis filterbank comprising a plurality of synthesis filters to a set of subband signals (215) derived from the set of output subband signals (213). [0121] EEE 15. The method (500) of EEE 14, wherein [0122] the synthesis filters comprise quadrature mirror filters; and/or [0123] a number of analysis filters is 32, 64 or more. [0124] EEE 16. The method (500) of any of EEEs 14 to 15, wherein the method (500) comprises [0125] upmixing the set of output subband signals (213) to a first and a second set of output subband signals (213) using one or more upmixing parameters; and [0126] deriving the set of subband signals (215) to which the synthesis filterbank is applied from the first set of output subband signals (213). [0127] EEE 17. The method (500) of any previous EEEs, wherein [0128] the input audio signal (110) exhibits an input sampling rate; [0129] the output audio signal (113) exhibits an output sampling rate; and [0130] the ratio of the output sampling rate and the input sampling rate is a rational number N/M; [0131] the anti-aliasing filter (102) comprises a set of coefficients; and [0132] performing (504) time domain resampling comprises multiplying a coefficient from the set of coefficients with a sample of the input audio signal (110) to determine a sample of the output audio signal (113). [0133] EEE 18. The method (500) of EEE 17, wherein [0134] the set of input subband signals (210) comprises Q input subband signals for Q different subbands ranging from 0 Hz to half of the input sampling rate; [0135] the method (500) comprises determining a number L of subbands to which a ripple pre-emphasis gain (323) is to be applied, based on the output sampling rate and the input sampling rate; and [0136] the method (500) comprises applying a ripple pre-emphasis gain (323) to L out of the Q input subband signals (210). [0137] EEE 19. The method (500) of EEE 18, wherein [0138] L is determined based on the ratio of the output sampling rate and the input sampling rate times Q, if the output sampling rate is smaller than the input sampling rate; and/or [0139] L is determined based on Q, if the output sampling rate is equal to or greater than the input sampling rate. [0140] EEE 20. The method (500) of any of EEEs 17 to 19, wherein performing (504) time domain resampling comprises, [0141] selecting a first subset of coefficients from the set of coefficients; wherein the first subset comprises a first coefficient of the set and coefficients of the set following the first coefficient by multiples of N; and [0142] determining a first sample of the output audio signal (113) based on the first subset of coefficients and a first plurality of samples of the input audio signal (110). [0143] EEE 21. The method of EEE 20, further comprising: [0144] selecting a second coefficient of the set based on the first coefficient and M; [0145] selecting a second subset of coefficients from the set of coefficients; wherein the second subset comprises the second coefficient and coefficients of the set following the second coefficient by multiples of N; and [0146] determining a second sample of the output audio signal (113) directly following the first sample, based on the second subset of coefficients and a second plurality of samples of the input audio signal (110). [0147] EEE 22. The method of any previous EEEs, wherein [0148] the frequency response (311) exhibits an amplitude ripple in a pass band range from 0 Hz to a Nyquist frequency of the input audio signal (110); [0149] the amplitude ripple exhibits R periods within the pass band range; [0150] the set of input subband signals (210) is determined using an analysis filterbank comprising Q analysis filters; [0151] R and Q are such that the ratio Q/R is equal to or greater than a pre-determined resolution value; [0152] the resolution value is 10, 20, 50 or more. [0153] EEE 23. The method (500) of any previous EEE, wherein the method (500) comprises [0154] determining a bandwidth of the input audio signal (110); and [0155] determining a number of input subband signals (210) from the set of input subband signals (210) to which a ripple pre-emphasis gain (323) is to be applied based on the bandwidth of the input audio signal (110). [0156] EEE 24. The method of any previous EEEs, wherein [0157] the set of input subband signals (210) is representative of a time domain audio signal at an input sampling rate; [0158] the input audio signal (110) exhibits the input sampling rate; and [0159] the output audio signal (113) exhibits an output sampling rate which differs from the input sampling rate. [0160] EEE 25. The method of any previous EEEs, wherein the method comprises [0161] determining a weighted deviation of the frequency response of the anti-aliasing filter (102) from a target frequency response, using a pass band weight for a pass band of the frequency responses and a stop band weight for a stop band of the frequency responses; and [0162] determining the anti-aliasing filter (102) such that the weighted deviation is reduced. [0163] EEE 26. The method of EEE 25, wherein the method comprises [0164] determining a stop band deviation of the frequency response of the anti-aliasing filter (102) from the target frequency response within the stop band; and [0165] determining the pass band weight and the stop band weight which are used for determining the anti-aliasing filter (102) such that the stop band deviation is equal to or smaller than a pre-determined deviation threshold. [0166] EEE 27. A system for determining a resampled output audio signal (113), wherein the system (200) comprises, [0167] a transform unit (203) configured to provide a set of input subband signals (210); [0168] a pre-emphasis unit (205) configured to apply a first ripple pre-emphasis gain (323) to a first input subband signal (210) of the set of input subband signals (210) to determine a corresponding first output subband signal (213) of a set of output subband signals (213); [0169] an inverse transform unit (208) configured to determine a time domain input audio signal (110) from the set of output subband signal (213); and [0170] a resampling unit (100) configured to perform time domain resampling of the input audio signal (110) to provide an output audio signal (113) using an anti-aliasing filter (102); wherein the first ripple pre-emphasis gain (323) is dependent on a frequency response (311) of the anti-aliasing filter (102). [0171] EEE 28. The system of EEE 27, wherein transform unit (203) and the inverse transform unit (208) form a perfect reconstruction filterbank or a near perfect reconstruction filterbank. [0172] EEE 29. An audio decoder (200) configured to provide an output signal (113) from a bitstream comprising audio data (221) and metadata (222), wherein [0173] the audio decoder (200) comprises the system of any of EEEs 27 to 28; and [0174] the transform unit (203) is configured to provide the set of input subband signals (210) based on the audio data (221). [0175] EEE 30. The audio decoder (200) of EEE 29, wherein the audio decoder (200) comprises a spectral band expansion unit (204) which is configured to provide at least some of the set of input subband signals (210) using the metadata (222). [0176] EEE 31. A storage medium comprising a software program adapted for execution on a processor and for performing the method steps of any of the EEEs 1 to 26 when carried out on a computing device.