JP2015228643A - Multi-band signal processor for digital audio signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow perceptual performance of a multi-band dynamic range compressor to be traded against computational resources in a flexible manner.SOLUTION: A method includes: processing an input signal to generate M delayed digital audio signal samples; converting the delayed digital audio signal samples to N frequency domain representations to compute respective signal spectrum values; determining respective signal level estimates; computing respective frequency domain gain coefficients on the basis of the respective signal level estimates and band gain laws; transforming the frequency domain gain coefficients to time domain representation to produce M time-varying filter coefficients of a processing filter; convolving the M delayed digital audio signal samples with the time-varying filter coefficients to produce a processed digital output signal; and updating the delayed digital audio signal samples in accordance with a sample rate or a predetermined block rate. In the method, two of the signal spectrum values for at least two of the frequency bands are updated at different rates, and M and N represent positive integers.

Description

  The present invention relates to multiband signal processors.

  Hearing instruments or hearing aids typically include a microphone amplification assembly that includes one or several microphones for receiving incoming sounds such as speech and music. This incoming sound is converted into an electrical microphone signal that is amplified and processed in the control and processing circuitry of the hearing aid device according to one or more preset listening programs. These listening programs have typically been calculated from user-specific hearing impairment or hearing loss, for example, expressed as an audiogram. Hearing aid output amplifiers can be used to process the processed microphone signal through a small speaker or receiver that can be housed in the hearing aid case with the microphone or separately in the earplug. To the ear canal.

  Hearing impaired people typically suffer a loss of hearing sensitivity, which depends on both the frequency and level of the sound of interest. That is, a hearing-impaired person can listen to a specific frequency (for example, low frequency) in the same manner as a normal hearing person, but at other frequencies (for example, high frequency), it has the same sensitivity as a non-deaf person. You may not be able to listen to the sound. Similarly, a hearing impaired person can perceive sounds, eg, loud sounds that exceed about 90 dB SPL, with the same sensitivity as a non-deaf person, but small sounds with the same sensitivity as a non-deaf person. You may not be able to listen. That is, in the latter situation, the hearing impaired suffers a loss of dynamic range at a specific frequency or frequency band. Various conventional analog and digital hearing aids have been designed to mitigate auditory defects with loss of dynamic range, as defined above. In order to compensate for the loss of dynamic range, prior art hearing aids use a so-called multi-band dynamic range compressor to compress the dynamic range of the incoming sound so that the compressed output signal is targeted. It is made to closely match the user's dynamic range. The ratio of the input dynamic range to the dynamic range output by the multiband dynamic range compressor is called the compression ratio. Multi-band dynamic range compressors typically have different compression ratios and / or in different conditions, such as different frequency bands, to accommodate the frequency-dependent loss of dynamic range of the target hearing impaired person. It is configured to operate with various attack time constants and release time constants.

  US Patent Application Publication No. 2003/0081804 discloses a so-called side-branch architecture for a multi-band dynamic range compressor based on Fast Fourier Transform (FFT). Multiband dynamic range compressors use side branches for frequency analysis of acoustic input signals. The FFT is calculated on the warp frequency scale from the exit tap of the cascade of first order allpass filters to which the acoustic input signal is applied. Similarly tapped delay lines are used for both FFT analysis and time-varying FIR compression filters. The result of the FFT-based frequency analysis is used to generate the coefficients of the FIR compression filter provided in the signal path.

  The warped frequency scale and side branch architecture of the disclosed multi-band dynamic range compressor has many desirable characteristics such as minimal time delay because the direct signal path includes only a short input buffer and FIR compression filter. can get. Other notable advantages are the absence of aliasing and natural logarithmic scaling of the analysis frequency band that fits well with the human auditory Bark based frequency scale. However, the disclosed FFT-based multiband dynamic range compressor has several undesirable characteristics. In particular, the signal spectral values for the entire frequency band of the FFT-based analysis are updated at the same block rate or frequency, which can lead to undersampling of the high frequency components of the input sound. Undersampling of the high frequency component is generally undesirable because it causes aliasing of the spectral level estimator in the analysis frequency band and leads to malfunction and distortion induction of the compression gain agent or coefficient.

  Furthermore, in a multiband dynamic range compressor based on FFT, a relatively high block rate may be selected to accommodate high frequency components, which is normal for the low frequency band of the analysis filter. This leads to faster updates than needed for correct sampling, ie oversampling in the low frequency band. The latter oversampling characteristic does not generate aliasing distortion, but consumes computational resources of the signal processor of the hearing aid to realize a multiband dynamic range compressor based on FFT. This process results in unnecessary power consumption by the hearing aid device, which shortens battery life.

  In view of the problems outlined above, an improved multi-band signal processor, such as a multi-band dynamic range compressor, that allows a separate and flexible update rate of the frequency band of the analysis filter is advantageous. Such an improved multiband signal processor would greatly increase the flexibility in selecting the block update rate for any particular frequency band of the analysis filter. Therefore, the sensing performance of the improved multi-band signal processor can be flexibly traded for the computational resources.

  A first aspect of the present disclosure includes a signal input for receiving a digital acoustic input signal, and M digital signals at each tapping node that receives the digital acoustic input signal and is interposed between the digital allpass filters. A multi-band signal processor comprising a cascade of digital all-pass filters configured to generate delayed digital acoustic signal samples. A multiband signal processor is a signal convolution processor configured to convolve M delayed digital acoustic signal samples with M time-varying filter coefficients of a processing filter to produce a processed digital output signal. (Signal convolution processor). The frequency domain transform processor is configured to transform the M delayed digital acoustic signal samples into a frequency domain representation to generate respective signal spectral values in a predetermined number N of frequency bands. The level estimator is configured to calculate each signal level estimator in a predetermined number of frequency bands based on each signal spectrum value. The processing gain calculator of the multiband signal processor is configured to calculate a frequency domain gain factor for each of a predetermined number of frequency bands based on the respective signal level estimators and band gain laws. ing. The inverse frequency domain transform processor is configured to transform the N frequency domain gain coefficients into M time-varying filter coefficients of the processing filter. The frequency domain transform processor is configured to calculate signal spectral values for at least two frequency bands at different band update rates. Each of M and N is a positive integer.

  The ability of the frequency domain transform processor to utilize different band update rates in at least two different frequency bands provides advantageous flexibility in selecting individual update rates for two or more of a predetermined number of frequency bands. This feature allows the perceptual performance of the multiband signal processor to be flexibly traded for computational resources. This feature also addresses and solves the above-mentioned problem of using the same update rate for all frequency bands imposed by prior art FFT-based processing. Having the same bandwidth update rate for all frequency bands means that a good update rate for the low frequency band usually gives a much higher update rate for the high frequency band than is required for proper sampling. It means to occur. Similarly, if an appropriate band update rate is selected for the low frequency band, the high frequency band will be undersampled, leading to aliasing and erroneous level estimators in the high frequency band. In contrast, the ability of the above frequency domain transform processor to apply individual band update rates to more than one of a predetermined number of frequency bands provides the optimum band update rate for each frequency band, while This means that aliasing distortion can be avoided and, on the other hand, oversampling and waste of computational resources can be avoided. The update rate for a particular band can also be optimized based on specific perceptual performance criteria of the multiband signal processor, such as speech intelligibility. In this way, the bandwidth update rate is relatively high in frequency bands that have a large impact on the perceptual performance criteria in question and relatively low in frequency bands that have a small impact on the perceptual performance criteria. May be. Accordingly, the computational resources of the frequency domain transform processor, level estimator and processing gain calculator may be assigned to those frequency bands that are important to the perceptual performance criteria.

  The multiband signal processor is preferably designed for hearing aid applications such that the number of delayed digital acoustic signal samples, M, is an even number between 8 and 64. This corresponds to M-1 digital allpass filters. The predetermined number of frequency bands, N, is preferably selected such that N = (M / 2) +1. In this embodiment, there is a single frequency domain gain factor for each frequency band generated by the frequency domain transform processor. In other words, there are a total of M time-varying filter coefficients to process M delay filter taps, whereas (M / 2) +1 of these M time-varying filter coefficients. Only is unique. The remaining (M / 2) -1 time-varying filter coefficients are determined from the fact that the (inverse) Fourier transform of the real gain vector yields a symmetric set of vector filter coefficients. Details of this conversion are described in US Patent Application Publication No. 2003/0081804.

  A person skilled in the art will set N = (M / 2) +1 so that the frequency domain transform processor applies a Discrete Fourier Transform (DFT) to calculate the signal spectrum value in the frequency band. It will be appreciated that it is particularly convenient when configured. However, the number of frequency bands, N, may be larger or smaller than (M / 2) +1, for example N = M / 2. In general, when the frequency domain transform processor (analysis filter) and the inverse frequency domain transform processor (synthesis filter) are correctly matched, an arbitrary number N ≦ M is set according to the requirements of the specific application of the multiband signal processor. May be used.

  The signal convolution processor of the multiband signal processor may update for each sample or update for the block. In the former case, the update rate of the signal convolution processor matches the sampling rate of the digital audio input signal, i.e. the reciprocal of the sampling frequency. The sampling frequency typically varies depending on the particular type of characteristics of the processing implemented by the multiband signal processor. The sampling frequency of the digital audio input signal is preferably between 16 kHz and 48 kHz in a multiband signal processor hearing aid application. If the signal folding processor is updated in blocks, each block may include a plurality of digital audio signal samples, eg, 4 to 64 digital audio signal samples. The band update rate for a particular frequency band specifies the frequency with which the signal spectrum value for that frequency band is calculated by the frequency domain transform processor. The maximum band update rate that may be applied to one frequency band, or a subset of frequency bands, of the predetermined number of frequency bands matches the update rate of the signal convolution processor. This maximum bandwidth update rate may be the sample rate or block rate of the signal convolution processor. When a signal spectrum value in a frequency band or a subset of frequency bands is calculated or updated, the corresponding signal level estimator and frequency domain gain factor are also preferably changed to the processing filter. To be reflected in the values of the M time-varying filter coefficients. On the other hand, recently calculated signal spectral values are maintained in the remaining frequency bands where the signal spectral values are not calculated or updated for a particular update or time step of the signal convolution processor. This also means that the corresponding signal level estimator and frequency domain gain factor are preferably maintained.

  As mentioned earlier, preferably the band update rate is generally adapted to the position of the frequency band so that the low frequency band has a lower band update rate than the high frequency band. The low frequency band may have a center frequency between 100 Hz and 500 Hz, for example, while the high frequency band may have a center frequency between 3 kHz and 8 kHz. Accordingly, one embodiment of a frequency domain transform processor calculates at least a first frequency band signal spectrum value at a first band update rate and at least a second frequency band signal spectrum value at a first frequency band. The update rate is lower than the bandwidth update rate, for example, 0.5, 0.33, or 0.25 times the first bandwidth update rate. The center frequency of the first frequency band is higher than the center frequency of the second frequency band.

  Those skilled in the art will recognize that the multi-band signal processor performs various signal processing of digital audio signals in many types of stationary and portable audio equipment, such as hearing aids, headsets, public loudspeakers, smartphones, tablets and others. It will be appreciated that it may be configured to perform a function. The multi-band signal processor is adapted to multi-band dynamic range compression of an acoustic input signal, multi-band dynamic range expansion of an acoustic input signal, acoustic input signal, etc. by appropriate design of one or more band gain rules of a processing gain calculator It may be configured to perform a signal processing function such as noise reduction.

  For each update of the signal convolution processor, the frequency domain transform processor may be configured to update respective signal spectral values of a predetermined number of frequency band subsets, and the level estimator may The signal level estimator of each subset may be configured to be updated, and the processing gain calculator updates each frequency domain gain factor of the frequency band subset and frequency domain gain of the residual frequency band. The inverse frequency domain transform processor may be configured to maintain the coefficients, and the inverse frequency domain transform processor transforms the updated frequency domain gain factor and the maintained frequency domain gain factor into M time-varying filter coefficients of the processing filter. You may comprise as follows.

  One skilled in the art will appreciate that the frequency domain transform processor may be configured to update the signal spectrum value for each frequency band of a predetermined number of frequency bands at a constant rate. This constant bandwidth update rate may be defined by a repeated bandwidth update schedule as described in more detail below. Despite a constant band update rate for each frequency band, there may be several subsets of frequency bands, where the band update rate is different between all frequency bands or has the same update rate. In another embodiment of the frequency domain transform processor, the band update rate for each frequency band is independently adapted based on the predicted need. The predicted need may be determined based on specific signal characteristics of the digital acoustic input signal, such as a predicted rate of change. The adaptive update rate may lead to further improvements in the computational load and performance tradeoff of the multiband signal processor.

  As described above, the frequency domain transform processor is preferably configured to update each signal spectrum value in a predetermined number of frequency bands according to a predetermined iterative band update schedule. The frequency domain transform processor may comprise a band selector that selects a particular frequency band that is updated in each sample update or block update of the convolution processor. The band selector thus recalculates or updates the signal spectrum values in any particular frequency band according to the band update schedule, in what order and thus at what update rate (how often) It may be controlled. One skilled in the art will understand that the update of the signal spectrum value in a particular band is preferably followed immediately by a corresponding update of the level estimate and frequency band gain factor for the frequency band of interest. The recurring bandwidth update schedule may be designed in a number of ways, for example using a so-called schedule matrix as described in more detail below with reference to the accompanying drawings. A person skilled in the art can use the frequency band update schedule to determine how often any particular frequency band is updated, i.e. the frequency of the frequency update, and the order in which the individual frequency bands are updated. It will be appreciated that significant flexibility is provided for the operation of the conversion processor. These features are known for improving the spectral coverage of the digital acoustic input signal, ie minimizing the modulation or gap in the time-frequency spectrum contained in the coverage of the band response. May be used to optimize the update rate of a particular frequency band.

  The signal processing functions of the multiband signal processor are implemented by dedicated digital hardware or implemented as one or more computer programs, routines and threads of execution running on a software programmable signal processor. Also good. Each of the computer program, routine, and thread of execution may include a plurality of executable program instructions. Alternatively, the signal processing functions may be realized by a combination of dedicated digital hardware and computer programs, routines and threads of execution running on a software programmable signal processor. For example, each of the aforementioned “frequency domain transform processor”, “signal convolution processor”, “inverse frequency domain transform processor”, “processing gain calculator” and “level estimator” includes a suitable microprocessor, particularly a digital A computer program, program routine, or thread of execution that may be executed on the signal processor may be included. The microprocessor and / or dedicated digital hardware may be integrated on the ASIC or implemented on the FPGA device.

  The frequency domain transform processor may be configured to calculate the signal spectral values of the M delayed digital acoustic signal samples in various ways without relying on the FFT algorithm. In a preferred embodiment of the frequency domain transform processor, the signal spectrum value of a single frequency band is calculated by vector-to-vector inner product using a discrete Fourier transform. This embodiment of the frequency domain transform processor is windowed or un-windowed of M delayed digital acoustic signal samples and rows of discrete Fourier transform matrices corresponding to frequency bands. ) As a vector dot product between discrete Fourier transform coefficients, each signal spectrum value of the frequency band is calculated.

  The inverse frequency domain transform processor is configured to transform the updated frequency domain gain factor and the maintained frequency domain gain factor into M time-varying filter coefficients by performing a set of scalar-vector multiplications. In this case, the scalar includes updated or maintained frequency domain gain coefficients, and the vector includes one row or one column of coefficients of the IFFT-based synthesis matrix.

  Specific signal processing functions implemented by the multiband signal processor may be conveniently defined by controlling the characteristics of the band gain rules. The bandwidth gain rule of the processing gain calculator may differ between different frequency bands. In one exemplary embodiment, the full-band gain rule may be configured to provide dynamic range compression of each signal in the frequency band, but certain compression parameters such as compression ratio and time constant are May vary between individual frequency bands. In one exemplary embodiment, the band gain rule is configured such that a first subset of a predetermined number of frequency bands is subjected to dynamic range compression, and another subset of frequency bands is a dynamic range extension. Alternatively, it may be different between different frequency bands so as to perform noise reduction or the like.

  Preferably, the one or more band gain rules of the processing gain calculator are one of multiband dynamic range compression of the acoustic input signal, multiband dynamic range expansion of the acoustic input signal, noise reduction of the acoustic input signal. Configured to do.

  As used herein, the term “band gain rule” refers to any function, relationship, equation, and / or algorithm that is configured to provide a particular feature associated with an acoustic input signal. The band gain rule may be arbitrarily defined depending on the embodiment.

  A second aspect of the present disclosure relates to a hearing aid device used by a user. The hearing aid is configured to generate a corresponding digital acoustic input signal coupled to the first microphone signal and a first microphone for generating a first microphone signal in response to receiving the sound. And a multi-band signal processor according to any of the above embodiments coupled or connected to a digital audio input signal. The multiband signal processor is configured to receive and process the first microphone signal in response to a user's hearing loss. The hearing aid device comprises a sound reproduction channel for receiving the processed digital output signal of the multi-band signal processor and converting it into an audible sound for transmission to the user.

  A third aspect of the present disclosure relates to a method of processing a digital audio input signal to produce a processed digital output signal, the method comprising: a) a digital audio input signal via a cascade of digital allpass filters. All-pass filtering to generate M delayed digital acoustic signal samples; b) transforming the M delayed digital acoustic signal samples into a frequency domain representation in a predetermined number N of frequency bands; Calculating a signal spectrum value; c) estimating each signal level in a predetermined number of frequency bands based on the signal spectrum value; and d) based on each signal level estimator and each band gain rule. , Each frequency domain gain factor for a given number of frequency bands E) transforming the frequency domain gain coefficient into a time domain representation to generate M time varying filter coefficients of the processing filter; and f) M delayed digital acoustic signal samples are processed into the processing filter. Convolution with the M time-varying filter coefficients to produce a processed digital output signal, and g) M delayed digital acoustics according to either the rate of each sample or a predetermined block rate Updating signal samples, wherein the signal spectral values of at least two different frequency bands are updated at different rates, each of M and N being a positive integer.

  According to a preferred embodiment of the method for processing a digital audio input signal, after each sample update of each of the M delayed digital audio signal samples, or after each block update: step b) comprises a subset of a predetermined number of frequency bands. Updating with the respective signal spectral values, step c) includes updating the respective signal level estimator of the frequency band subset, and step d) comprises updating each of the frequency band subsets. Updating the frequency domain gain factor and maintaining the previous frequency domain gain factor of the remaining frequency band, and step e) processing the updated frequency domain gain factor and the maintained frequency domain gain factor Converting to an updated value of the M time-varying filter coefficients of the filter.

  One skilled in the art will appreciate that a subset of frequency bands may include only a single frequency band. In the latter embodiment, the signal spectrum value for a single frequency band is calculated for each execution of step f), where M delayed digital acoustic signal samples are convolved with M time-varying filter coefficients of the processing filter. Is updated. This is particularly advantageous when the convolution processor operates in the mode of each sample discussed above, because this mode is determined by the preferred design of the bandwidth update schedule discussed above. This is to allow the frequency band to have a high update rate.

  Preferably, different subsets of frequency bands are updated during successive sample updates of M delayed digital acoustic signal samples or during successive block updates according to a predetermined iterative band update schedule. .

  A fourth aspect of the present disclosure is configured to cause a signal processor to perform steps a) -g) of the method for processing a digital acoustic input signal as outlined above to produce a processed digital output signal. Relates to a computer readable data carrier containing executable program instructions. The computer readable data carrier may comprise a magnetic disk, an optical disk, a memory stick or any other suitable data storage medium.

  The multiband signal processor has a signal input for receiving a digital acoustic input signal; M delayed digital acoustic signals at each tapping node that receives the digital acoustic input signal and is interposed between the digital all-pass filters. A cascade of digital all-pass filters configured to generate samples; convolution of the M delayed digital acoustic signal samples with the M time-varying filter coefficients of the processing filter, and the processed digital output A signal convolution processor configured to generate a signal; configured to convert the M delayed digital acoustic signal samples into a frequency domain representation to generate respective signal spectral values in the N frequency bands. A frequency domain transform processor; A level estimator configured to calculate respective signal level estimators in N frequency bands based on respective signal spectral values; and based on respective signal level estimators and band gain rules, A processing gain calculator configured to calculate a frequency domain gain factor for each of the N frequency bands; and to convert the N frequency domain gain factors into M time-varying filter coefficients of the processing filter. An inverse frequency domain transform processor configured to provide at least two of the signal spectrum values at least two of the frequency bands at different band update rates, and M is It is a positive integer and N is a positive integer.

  Optionally, the signal convolution processor is configured to be updated for each sample or to be updated for each block, where each block includes a plurality of digital acoustic signal samples.

  Optionally, the frequency domain transform processor calculates one of the signal spectral values for the first frequency band of the N frequency bands at the first band update rate, and is greater than the first band update rate. Configured to calculate another one of the signal spectral values for the second of the N frequency bands at a low update rate, wherein the center frequency of the first frequency band is the second frequency band It is higher than the center frequency of the frequency band.

  Optionally, the signal convolution processor is updated in multiple updates, for each update: the frequency domain transform processor updates a subset of signal spectral values for a subset of N frequency bands. The level estimator is configured to update a subset of the signal level estimator for a subset of the N frequency bands; and the processing gain calculator is a portion of the N frequency bands. The frequency domain gain coefficient subset is updated for the set, and the remainder of the frequency domain gain coefficient is maintained for the remainder of the N frequency bands.

  Optionally, the subset of frequency bands is formed by a single frequency band of N frequency bands.

  Optionally, the inverse frequency domain transform processor transforms the updated frequency domain gain factor and the maintained frequency domain gain factor into M time-varying filter coefficients by performing a set of scalar-vector multiplications. The scalar included in the scalar-vector multiplication includes an updated frequency domain gain factor or a maintained frequency domain gain factor, and the vector included in the scalar-vector multiplication is an IFFT-based composite matrix Contains one row or column of coefficients.

  Optionally, the frequency domain transform processor is configured to update the signal spectral values for each frequency band at a constant update rate.

  Optionally, the frequency domain transform processor is configured to update the signal spectrum values according to a predetermined iterative band update schedule.

  Optionally, the frequency domain transform processor includes windowed or un-windowed discrete Fourier transform coefficients of the M delayed digital acoustic signal samples and the rows of the discrete Fourier transform matrix. Is configured to calculate at least one of the signal spectral values as a vector dot product between.

  Optionally, one or more of the band gain rules are configured to provide multi-band dynamic range compression of the digital audio input signal, multi-band dynamic range expansion of the digital audio input signal, or noise reduction of the digital audio input signal Has been.

A hearing aid device for use by a user, comprising: a multi-band signal processor; a first microphone coupled to the multi-band signal processor; and a speaker coupled to the multi-band signal processor. .
A method for processing a digital acoustic input signal to produce a processed digital output signal is disclosed. The method includes the steps of all-pass filtering the digital audio input signal through a cascade of digital all-pass filters to generate M delayed digital audio signal samples; Converting to a frequency domain representation in the number of frequency bands and calculating a respective signal spectrum value; determining a respective signal level estimator in the N frequency bands based on the signal spectrum value; Calculating respective frequency domain gain coefficients for the N frequency bands based on the signal level estimator and the band gain rule; and converting the frequency domain gain coefficients into a time domain representation to generate M processing filters. Generating a time-varying filter coefficient of Convolution of the M delayed digital acoustic signal samples with the M time-varying filter coefficients of the processing filter to produce a processed digital output signal; according to the rate of each sample or a predetermined block rate Updating M delayed digital acoustic signal samples, wherein at least two of the signal spectral values for at least two of the N frequency bands are updated at different rates, and M is a positive integer. , N are also positive integers.

  Optionally, the method updates a subset of signal spectral values for a subset of N frequency bands; updating a subset of signal level estimators for a subset of N frequency bands; Updating a subset of frequency domain gain factors for a subset of N frequency bands; and maintaining a remainder of frequency domain gain factors for the remainder of N frequency bands.

  Optionally, the M delayed digital acoustic signal samples are updated according to a predetermined repetitive band update schedule.

  Disclosed is a computer product comprising a non-transitory medium that stores executable program instructions that perform any of the foregoing methods by performing it with a signal processor.

  Other and further aspects and features will become apparent from the following detailed description of the embodiments.

  Embodiments will be described in more detail in connection with the following accompanying drawings.

1 is a schematic block diagram of a prior art Fast Fourier Transform (FFT) based multi-band dynamic range compressor having a side branch architecture. FIG. FIG. 2 is a simplified schematic block diagram of a multi-band dynamic range compressor, according to some embodiments. FIG. 3 is a schematic block diagram illustrating calculation of a set of time-varying compression filter coefficients of the compression filter of the multi-band dynamic range compressor of FIG. 2. FIG. 3 shows a first exemplary band update schedule based on a matrix filling method for the multi-band dynamic range compressor of FIG. FIG. 3 shows a second band update schedule for the multi-band dynamic range compressor of FIG. 2 including a frequency band repeat pattern. 4B is a time-frequency plot of the processed output signal of a multi-band dynamic range compressor using the band update schedule shown in FIG. 4B to optimize spectral coverage. FIG.

  Various embodiments will be described below with reference to the drawings. It should be noted that the drawings are not necessarily drawn to scale, and elements of similar structure or function are represented by the same reference numerals throughout the drawings. Moreover, the drawings are only intended to facilitate the description of the embodiments. These drawings are not intended to be an exhaustive description of the invention recited in the claims, nor are they intended to limit the invention as described in the claims. Moreover, the illustrated embodiments need not have all the aspects or advantages shown. The aspects or advantages described in connection with a particular embodiment are not necessarily limited to that embodiment, even if not so illustrated or explicitly described as such. It can be implemented in any other embodiment.

FIG. 1 is a schematic block diagram of a multi-band dynamic range compressor 100 based on a prior art Fast Fourier Transform (FFT) having a so-called side branch architecture. The multiband dynamic range compressor 100 uses a side branch to perform frequency analysis, compression gain coefficient calculation, and frequency synthesis of an acoustic input signal. The digital acoustic input signal x (n) is applied to the input 1001 of the multiband dynamic range compressor 100 and propagates through a cascade of K first-order allpass filters A (z) to produce a delayed digital acoustic signal. A sequence of samples p ₀ (n) to p _K (n) is generated. By using a first-order all-pass filter rather than the traditional pure delay, the frequency scale on which frequency analysis and frequency synthesis are performed becomes a so-called warped frequency scale with many desirable characteristics as discussed above. Convert. The sequence of delayed samples p ₀ (n) -p _K (n) is then windowed and an FFT is calculated using the windowed sequence (1005). The result of the FFT is a frequency spectrum sampled at regular intervals on the Bark frequency scale. Since the input data sequence is windowed, the frequency spectrum is smoothed in the warped frequency domain, thereby generating overlapping frequency bands. A frequency domain level estimator (eg, power spectrum) is calculated from the warp FFT, and then a frequency domain gain factor (eg, compression gain) is calculated from the warp power spectrum for the acoustic analysis band. (1007). Since the frequency domain gain factor is a pure real number, the result of the inverse FFT of the warped time domain filter is a real number and a set of filter coefficients with even symmetry is obtained (1009). The system sound output y (n) is then calculated by convolving the delayed digital sound signal sample sequence p ₀ (n) -p _K (n) with a compression gain filter (1011), in this case g _K (N) is a compression filter coefficient. Operating at 1.5 ms for 24 samples per block at a given block rate, eg 16 kHz sampling frequency, the FFT operation 1005 is a synchronous update of the power spectrum in all frequency bands Therefore, it leads to the same update rate for all frequency bands, leading to the problems discussed above.

  FIG. 2 is a simplified schematic block diagram of a multi-band signal processor 200 according to some embodiments. In this embodiment, the multi-band signal processor is configured to function as a multi-band dynamic range compressor 200, but those skilled in the art will appropriately adapt in other embodiments of the multi-band signal processor. It will be appreciated that other signal processing functions such as multi-band expansion or noise reduction may be implemented.

  Multiband dynamic range compressor 200 includes a signal input, Audio in, for accepting a digital audio input signal to compressor 200. A direct acoustic signal path through the compressor 200 receives the digital acoustic input signal and generates a plurality of delayed digital acoustic signal samples at each tapping node (spot) interposed between the digital all-pass filters. M-1 digital all-pass filters 201a, 201b,. . . 201M-1 cascade. Cascade digital all-pass filters 201a, 201b,. . . The number of 201M-1 will vary depending on the performance and power requirements of the particular application of the compressor 200. In a number of useful embodiments for multi-band compression in hearing aid applications, the number of digital allpass filters, M-1, is 7 to 63, and 8 to 64 delayed digital acoustic signal samples, respectively, Generate at the tapping node.

The direct acoustic signal path includes M multipliers 202a,. . . Further included is a signal convolution processor that includes a summing function 215 coupled to the 202 outputs of 202M. The signal convolution processor convolves a contiguous sample, or block of samples, of a plurality of delayed digital acoustic signal samples with a predetermined update rate using _M time-varying compression filter coefficients g ₁ -g _M of the compression filter. Arithmetic and configured to generate a processed digital output signal in the digital audio output, Audio Out (n), of the multi-band compressor 200. N is a positive integer and is preferably less than or equal to M. One skilled in the art will appreciate that the update rate may vary depending on whether the processing of the multi-band compressor 200 is based on blocks or on each sample. . In an embodiment based on the block of multi-band compressor 200, the block of samples may include all or any subset of the M delayed digital acoustic signal samples at the tapping node. The sample-by-sample update rate of the multi-band compressor 200 allows the multi-band compressor 200 to respond particularly quickly to impulse noise or other undesirable transients in the digital acoustic input signal. , Thereby minimizing user discomfort. A person skilled in the art k = 1, 2,. . . The tapping node 202k for M performs the multiplication x_k (n) * g_k (n), where “*” means multiplication, and x_k (n) is the delay line, or digital allpass, at time n. It will be understood that the signal at the kth tap of the cascade of filters, g_k (n) is the kth time-varying compression filter coefficient at time n. The adder node 215 simply adds the inputs, outputs the result, and sends it to Audio Out (n). The signal convolution processor performs the following calculations:
Audio Out (n) = sum_k [x_k (n) * g_k (n)], where sum_k means the total from k = 1 to M.

The multi-band compressor 200 further includes a so-called side chain processor or side chain function 205, which is a frequency domain transform processor 203 (often referred to as “analysis filter bank”), an inverse frequency domain transform processor. 209 (often referred to as “synthesis filter bank”) and a processing gain calculator 207 interposed between the two transform processors 203, 209. Lastly, the sidechain processor or sidechain function 205 includes a band selector 206, which, as will be described in more detail below with reference to FIG. Controls the order, i.e., frequency, in which signal spectral values and associated frequency domain gain factors in a particular frequency band are calculated or updated. The output of the sidechain processor 205 is the N time-varying compression filter coefficients or compression vectors, g _{1 to} g _M , discussed above. The M delayed digital acoustic signal samples are added to the input of a frequency domain transform processor 203, which transforms the delayed digital acoustic signal samples into a frequency domain representation for a predetermined number of times created by the transformation process. A signal spectrum value in each frequency band is generated. The number of frequency bands may match M / 2 + 1, and the setting of M = 32 may correspond to 17 frequency bands. These frequency bands preferably overlap with an overlap controlled by a window function, eg, a Hanning window.

The frequency domain transform processor 203, or alternatively the processing gain calculator 207, is configured to calculate a signal level estimator for each of the frequency bands based on a predetermined signal spectrum value in the frequency band of interest. A level estimator (not shown in FIG. 2, but shown as item 313 in FIG. 3) is provided. The signal level estimator may include, for example, an amplitude, power or energy level estimator. The level estimator at each frequency may also include specific time constants such as attack time and release time. The band gain rule defines, for example, a specific compression ratio of an acoustic signal within a target frequency band. The compression ratio may be constant over the entire level of the digital sound input signal or may be variable over the dynamic range of the digital sound input signal. Band gain rules may be defined in various ways. In one embodiment, the band gain rule may be defined via a look-up table that maps signal level estimator values to corresponding frequency domain gain factor G _k values. Band gain rules may differ between different frequency bands or may be essentially the same in more than one frequency band. However, it may be advantageous to use different band gain rules and thus different compression parameters in at least two different frequency bands in order to provide optimal hearing loss compensation for hearing loss of hearing impaired users. Many.

Calculated frequency-domain gain factor G _k is the coefficient synthesis described below, a frequency-domain gain factor, which is configured to convert the M time compression factor filter change compression filter coefficients g ₁ to g _M , And sent to the inverse frequency domain transform processor 209.

  As described above, the band selector 206 controls the order in which signal spectrum values are calculated or updated in any particular frequency band of the plurality of individual frequency bands, and thus controls its frequency. Preferably, the band update rates of the signal spectrum values of at least two different frequency bands are different. For example, by making the band update rate in the low frequency band centered at 200 Hz, for example, generally lower than the band update rate in the high frequency band centered at about 5 kHz, the signal level in the high frequency band can be accurately More computational resources may be allocated for estimation. This is advantageous because, for example, it is expected that the level of the incoming sound will change more quickly for what is caused by the impact noise discussed above. The band update rate of the high frequency band may be equal to, for example, the block rate of the convolution processor of the multiband compressor 200 or the update rate of each sample. The update rate of each sample corresponds to the inverse of the selected sampling frequency of the digital acoustic input signal. This sampling frequency may be 16 kHz to 48 kHz for typical hearing device applications. A block of samples may include 4 to 64 samples. On the other hand, the low-frequency band update rate corresponds to each sample of the second, third, fourth, etc. or each block of the second, third, fourth, etc. The update rate of the high frequency band may be smaller than at least one half. This difference in update band rates between different frequency bands is in contrast to the prior art multi-band dynamic range compressor 100 based on the Fast Fourier Transform (FFT) discussed above. In the prior art, because of block processing based on FFT, the update rate of the power spectrum in the entire frequency band is the same.

FIG. 3 is a schematic block diagram illustrating in more detail how the _M time-varying compression filter coefficients g ₁ -g _M discussed above are calculated in the side chain function 205 of FIG. 2 or the branch of the multi-band compressor 200. FIG. M delayed digital acoustic signal samples x _{n to} x _{n −M + 1} are applied to the input of each multiplier of the frequency domain transform processor 203. The conversion to the frequency domain representation is performed in this embodiment by a short time Fourier transform (STFT) algorithm, and the weighting of the STFT coefficients is obtained as a result of windowing with an appropriate window function, for example a Hanning window. It is done. This calculation, matrix - a vector product, the product of the M delay digital audio signal samples x _n ~x _{n-M + 1} and STFT coefficient matrix, vector - using the characteristics that can be executed as a sequence of the vector product Yes. More specifically, the STFT matrix of M coefficients in the kth row of STFT coefficients, denoted W _{1k to} W _Mk in the detailed flowchart 203a, and M delayed digital acoustic signal samples, The inner product between them determines the updated signal spectrum value in the k th frequency band of the analysis filter 203. The updated signal spectrum value of the kth frequency band appears at the output of the addition function 311. The signal level estimation amount P _k of the signal spectrum value in the k-th frequency band is obtained by the power estimation function 313, in which the logarithmic power estimation amount squares the signal spectrum value and the logarithm of the result is obtained. Formed by taking. The signal level estimator P _k is applied to the processing gain calculator 207 discussed above, which calculates the frequency domain gain factor G _k based on the previously considered band gain rule for the k th frequency band. Calculate the corresponding update value. On the other hand, each previous value of the frequency domain gain factor for the remaining N-1 frequency bands that remains unprocessed or not updated in the current sample period is the side chain processor or side chain function. Maintained by 205 appropriate memory elements (not shown). The vector dot product between the STFT coefficient and the M delayed digital acoustic signal samples is a specific row of STFT coefficients, eg, W _{19 to} W _M9 and W corresponding to the 9th and 13th frequency bands, respectively. It is implemented for each of the preselected sets of STFT vectors, including _{13 to} W _M3 , respectively. In this way, signal spectrum values in each frequency band are calculated in a predetermined order or sequence controlled by the band selector 206. This order or order is a predetermined band update schedule or band sampling schedule that determines in what order and how often the signal spectrum values of any particular frequency band are calculated or updated. Define

  An exemplary bandwidth update schedule or sampling schedule 400 is illustrated in FIG. 4A. The example of FIG. 4A is for the multi-band compressor 200 described above in an embodiment where M = 32 and 17 different frequency bands. Further, in this embodiment, the signal convolution processor is updated at the rate of each sample, and the characteristics of the exemplary band sampling schedule 400 are adapted to the update rate of each sample. One skilled in the art will appreciate that the signal convolution processor in alternative embodiments may be updated at an appropriate block rate to adapt the bandwidth update schedule thereto.

  Band sampling schedule 400 is arranged as a schedule matrix 400 containing 8 columns and 6 rows of integers, which are frequency band indexes that define specific frequency bands of 17 different frequency bands. In updating each sample of the signal convolution processor of the multi-band compressor, only a single frequency band is processed and updated. The number of frequency bands that are updated or processed at any particular sample time is indicated by a “band update schedule”. Next, consider a column wise trajectory passing through the schedule matrix 400, as shown by the closed loop schedule curve 402. This trajectory defines the bandwidth update schedule. In each sample period, band selector 206 proceeds to the next entry in the band update schedule, thereby selecting the frequency band to be updated or processed. Accordingly, the direction of the closed loop schedule curve 402 indicates the direction and order in which the frequency bands are updated / sampled, and the respective level estimators and associated frequency domain gain factors of the frequency bands are updated. For example, if the closed loop schedule curve 402 is located at matrix entry (3, 7) at sample time step n, the frequency band selected will be number 13 as shown by circle 405 in schedule matrix 400. It is. In the subsequent sample time step (n + 1), the closed loop schedule curve becomes a matrix entry (4, 7) and the frequency band number 15 is updated.

  The closed nature of the closed loop schedule curve 402 means that the bandwidth update schedule or sampling schedule is repeated every 6 * 8 = 48 sample periods or time steps. Note that, depending on the construction, this matrix filling method is such that the number of sample time periods or steps between selection and update of the same frequency band, i.e., the band update rate, or the band sampling period, is constant for each frequency band. To be certain. For example, the band update rate for frequency band 4 is 48 sample periods, corresponding to a complete cycle through schedule matrix 400. For a 16 kHz sampling frequency of the digital acoustic input signal, this corresponds to a band update rate of about 3 ms. For frequency band 9, the band update rate is 24 sample periods and for frequency band 16, the band update rate is 6 sample periods, which are about 1.5 ms and about 0.375 ms, respectively. Correspond.

  Accordingly, the band update rates for frequency bands 4, 9 and 16 are all different in this particular embodiment. In other words, the novel feature of creating a valid bandwidth update schedule using the schedule matrix 400 method is that, by construction, the bandwidth update rate is constant for any given bandwidth number. In some cases, it is unique. Updates to any particular frequency band can be increased if the band selector 206 is configured to repeat the pattern more frequently when it “fills” the schedule matrix 400. One skilled in the art will appreciate that the bandwidth update schedule may be constructed in a number of ways and adapted to the requirements of a particular application. A band schedule is, for example, such that each frequency band has its own unique update rate / sampling period, ie, the two frequency bands are never scheduled at the same time, and the update rate of each frequency band is constant. It may be constructed to remain. One method of constructing a bandwidth schedule by the “matrix filling method” is illustrated by the schedule matrix 400. Numbers 1 to 8 are arranged in the first row. Numbers 9 to 12 are arranged in the second row, numbers 13 and 14 are arranged in the third row, and the fourth to sixth rows hold the numbers 15 to 17. Filled matrix entries are shaded with a gray background. Row 1 is completely filled, while rows 2 through 6 are only partially filled. By the way, the empty places in each line may be filled by repeating the first number pattern in each line. For example, row 3 defines frequency bands 13, 14, 13, 14, 13, 14, 13, and 14, that is, patterns 13 and 14 are repeated four times. Thus, the schedule matrix 400 is filled with (multiple) frequency band indexes 1 to 17.

  Those skilled in the art will recognize that many variations on the matrix filling method described above are possible. For example, differently sized matrices can be filled with numbers 1 to 17, resulting in different bandwidth update schedules. Alternatively, for a filter bank having 17 frequency bands, the matrix can be filled with a number greater than 17, eg, numbers 1-20. If a frequency band number greater than 17 is selected, no available frequency band is updated or processed. This scheme results in a filter bank in which either a single band is processed or no frequency band is processed in each sample time period. Such a bandwidth update schedule is advantageous because it allows the battery consumption of the multi-band compressor to be traded off for perceptual performance. In the same sense, schedule matrix 400 can also describe possible band update schedules for any filter bank having a frequency band of less than 17. Finally, it is possible to perform permutation on the bandwidth schedule matrix 400. For example, if frequency band numbers 8 and 4 are exchanged in the bandwidth schedule matrix 400, the effective bandwidth update is different. You will get a schedule. In addition, ad hoc modification to the band schedule matrix 400 is also possible to ensure that an effective schedule is obtained. An important reason for permutation of the band update schedule is to improve the spectrum coverage, i.e., minimize the modulation or gap in the time-frequency spectrum included in the coverage by the band response. An example of the optimized spectral coverage of the band sampling schedule, along with the corresponding (9 × 4) schedule matrix 400a shown in FIG. 4B, is shown in the time-frequency plot of FIG.

  Further, the exemplary band sampling schedule 400 is for at least some low frequency bands such as bands 1, 2, 3, 4 and 5 than for high frequency bands such as bands 14, 15 and 16. It is clear to define a lower bandwidth update rate. In addition, the design space without constraints on the band sampling schedule greatly increases the flexibility of when to update signal spectral values and corresponding level estimates and frequency domain gain factors for specific frequency bands. . Thus, the individual bandwidth update rates of the frequency bands can greatly reduce the computational load for perceptual performance equivalent to the prior art FFT-based multiband dynamic range compressor 100 discussed above. . In an improved embodiment, the band update rate for each frequency band can be adapted independently based on the predicted need.

The compression gain filter is preferably configured so that the selected frequency band is a band sampling so that all the updated values of the _M time-varying compression filter coefficients g _{1 to} g _M are calculated for each sample time period. • Updated when updated according to schedule. In order to apply the compression gain filter in the time domain, that is, obtain updated values of _M time-varying compression filter coefficients g _{1 to} g M that reflect the updated value of the frequency domain gain coefficient G _k of the k′-th band. Therefore, it is necessary to convert back to the time domain. There are in principle at least two different ways in which this can be done: (1) perform an IFFT on the frequency domain gain factor G _k and with an appropriate synthesis window element-wise Matrix to perform multiplication or (2) proceed as shown in detailed flow diagram 209a of inverse frequency domain transform processor 209 to combine the matrix with the appropriate IFFT based vector using an appropriate synthesis window -Use vector multiplication. Using the latter method, taking into account that only one frequency band is updated every sample time period, the compression filter coefficients g ₁ -g _M or coefficient vectors are updated incrementally by scalar-vector multiplication. in this case, the scalar is the frequency-domain gain factor of interest, namely the like G _k for the k 'th frequency band, vector is one row or one column of coefficients from the synthesis matrix based on IFFT. When proceeding in this manner, the frequency domain gain factor G _k is first applied to the exponential function 315 to transform G _k from the log domain (created by the log function 313) to a linear representation. . In a subsequent step, the previous value of the frequency domain gain factor G _k for the selected k ′ th frequency band is subtracted from the current value of G _k by the subtractor 317 to obtain the value of the frequency domain gain factor G _k . An update or increment occurs. The frequency domain gain factor value increment is then multiplied by the weighted inverse Fourier coefficients from the IFFT-based synthesis matrix, shown as coefficients V _1k -V _Mk on detailed flow diagram 209a. This step converts the calculated increment in the value of the frequency domain gain factor into a corresponding increment or update of the compression filter coefficients g _{1 to} g _M. The increment or update of the compression filter coefficients g _{1 to} g _M corresponds to the update of the frequency domain gain coefficient of the k ′ th frequency band. These filter coefficients increments, which are connected to each compression filter coefficients g ₁ to g _M, memory / delay function and addition function 319a, the 319b, are added to each of the previous filter coefficients. Therefore, the values of the compression filter coefficients g _{1 to} g _M are updated. One skilled in the art will recognize that performing the latter conversion in fixed point arithmetic may result in accumulation of rounding errors. Therefore, it is sometimes preferred to replace the synthesis scheme outlined above with a full-band inverse discrete Fourier transform. The latter resets or eliminates the accumulated rounding error.

FIG. 5 is a time-frequency plot showing the optimized spectral coverage achieved by employing the schedule matrix 400a shown in FIG. 4B in the 17 bands of the multi-band compressor 200. 500. The frequency axis is linear from 0 Hz to 8000 Hz, which corresponds to the Nyquist frequency of the used 16 kHz sampling rate of the acoustic input signal to the multi-band compressor 200. The y-axis shows the time in the sample of the acoustic signal, where 100 samples correspond to 6.25 ms. The shade scale in the time-frequency plot indicates the relative spectral coverage of the filter bank. The white / light color (corresponding to about 1.0 on scale 502) at a given time-frequency coordinate in the plot allows the filter bank to resolve the time-frequency position in the spectrum very well. Indicates. In this context, “resolve well” means that any acoustic input signal at the spectral position of interest is properly measured or detected by the spectral power estimator P _k . On the other hand, the black color on the scale 502 (corresponding to about 0.0-0.2) is the band power estimation process in which the acoustic input signal at that time-frequency position is performed in the side-chain processor of the multi-band compressor. Indicates that it remains almost undetected. Each location of the 17 frequency bands on the time-frequency plot is indicated by a corresponding band index. It is clear that the selected band sampling schedule defined by schedule matrix 400a provides good spectral coverage over a frequency range of about 100 Hz to 4 kHz which is important for speech comprehension. This is indicated by the relative white or light gray over time in the frequency band 1-13. The resolution for frequency bands 14-17 is small or bad, but these frequency bands are less important for speech comprehension. It is up to the designer of the band sampling schedule to determine whether the spectral coverage characteristics will suit the application at hand. By changing the band sampling schedule, the designer can influence the characteristics of the spectral coverage pattern in a very straightforward and highly flexible manner as described above.

  While specific embodiments have been shown and described, it is not intended that the invention described in the claims be limited to the preferred embodiments, and those skilled in the art will It will be apparent that various changes and modifications can be made without departing from the spirit and scope of the invention. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. The claimed invention is intended to include within its scope alternatives, modifications and equivalents.

  The present disclosure includes a plurality of aspects according to the items described below.

(Item 1)
A multi-band signal processor,
A signal input for receiving a digital acoustic input signal;
A cascade of digital allpass filters configured to receive the digital acoustic input signal and to generate M delayed digital acoustic signal samples at each tapping node interposed between the digital allpass filters;
A signal convolution processor configured to convolve M delayed digital acoustic signal samples with the M time-varying filter coefficients of the processing filter to produce a processed digital output signal;
A frequency domain transform processor configured to convert the M delayed digital acoustic signal samples into a frequency domain representation to provide respective signal spectral values in a predetermined number N of frequency bands;
A level estimator configured to calculate respective signal level estimators in a predetermined number of frequency bands based on respective signal spectral values;
A processing gain calculator configured to calculate a frequency domain gain factor for each of a predetermined number of frequency bands based on respective signal level estimators and band gain rules;
An inverse frequency domain transform processor configured to convert N frequency domain gain factors to M time varying filter coefficients of the processing filter, the frequency domain transform processor at least with different band update rates A multiband signal processor configured to calculate signal spectrum values for two frequency bands, each of M and N being a positive integer.

(Item 2)
The multiband signal processor of item 1, wherein the signal convolution processor is configured to be updated for each sample, or configured to be updated for a block, wherein each block includes a plurality of digital acoustic signal samples.

(Item 3)
The frequency domain transform processor
Calculating a signal spectrum value of at least a first frequency band at a first band update rate;
Configured to calculate a signal spectrum value of at least a second frequency band at an update rate lower than the first band update rate;
Item 3. The multiband signal processor according to item 1 or 2, wherein the center frequency of the first frequency band is higher than the center frequency of the second frequency band.

(Item 4)
In each update of the signal convolution processor:
The frequency domain transform processor is configured to update a respective signal spectral value of a subset of the predetermined number of frequency bands;
The level estimator is configured to update each signal level estimator of the subset of frequency bands,
The processing gain calculator is configured to update each frequency domain gain factor of the subset of frequency bands and maintain the frequency domain gain factor of the remaining frequency bands,
The inverse frequency domain transform processor of claim 2, wherein the inverse frequency domain transform processor is configured to transform the updated frequency domain gain factor and the maintained frequency domain gain factor into M time-varying filter coefficients of the processing filter. Band signal processor.

(Item 5)
Item 5. The multiband signal processor of item 4, wherein the subset of frequency bands is formed by a single frequency band of a predetermined number of frequency bands.

(Item 6)
6. The multiband signal processor according to any of items 1-5, wherein the frequency domain transform processor is configured to update the signal spectrum value of each frequency band at a constant update rate.

(Item 7)
Item 7. The multiband signal processor of item 6, wherein the frequency domain transform processor is configured to update respective signal spectrum values in a predetermined number of frequency bands according to a predetermined iterative band update schedule.

(Item 8)
A frequency domain transform processor windowed or un-windowed discrete Fourier transforms of M delayed digital acoustic signal samples and rows of discrete Fourier transform matrices corresponding to frequency bands 8. A multiband signal processor according to any of items 1 to 7, configured to calculate respective signal spectral values of the frequency band as vector dot products between the coefficients.

(Item 9)
An inverse frequency domain transform processor is configured to convert the updated frequency domain gain factor and the maintained frequency domain gain factor into M time-varying filter coefficients by performing a set of scalar-vector multiplications. And
Item 5. The multiband signal processor of item 4, wherein the scalar includes an updated or maintained frequency domain gain factor, and the vector includes one row or column of coefficients of a synthesis matrix based on IFFT.

(Item 10)
One or more of the bandwidth gain rules of the processing gain calculator are:
Multi-band dynamic range compression of digital audio input signal,
Multi-band dynamic range expansion of digital sound input signal,
Item 10. The multiband signal processor of any of items 1-9, wherein the multiband signal processor is configured to provide one of noise reduction of a digital acoustic input signal.

(Item 11)
A hearing aid used by a user,
A first microphone for generating a first microphone signal in response to receiving the sound;
An acoustic input channel coupled to the first microphone signal and configured to generate a corresponding digital acoustic input signal;
A multi-band signal processor according to any of items 1-10, coupled to a digital acoustic input signal, configured to receive a first microphone signal and process in response to a hearing loss of a user;
A hearing aid device comprising: a sound reproduction channel for receiving the processed digital output signal and converting it into an audible sound for delivery to a user.

(Item 12)
A method of processing a digital acoustic input signal to produce a processed digital output signal comprising:
a) allpass filtering the digital acoustic input signal through a cascade of digital allpass filters to generate M delayed digital acoustic signal samples;
b) converting the M delayed digital acoustic signal samples into a frequency domain representation in a predetermined number N of frequency bands and calculating respective signal spectral values;
c) estimating each signal level in a predetermined number of frequency bands based on the signal spectrum values;
d) calculating respective frequency domain gain factors for a predetermined number of frequency bands based on respective signal level estimators and respective band gain rules;
e) converting the frequency domain gain factor to a time domain representation to generate M time varying filter coefficients of the processing filter;
f) convolution of the M delayed digital acoustic signal samples with the M time-varying filter coefficients of the processing filter to produce a processed digital output signal;
g) updating M delayed digital acoustic signal samples according to either the rate of each sample or a predetermined block rate, wherein the signal spectral values of at least two different frequency bands are updated at different rates And each of M and N is a positive integer.

(Item 13)
After updating for each sample of the M delayed digital acoustic signal samples or after updating for each block:
Step b) comprises updating a subset of the predetermined number of frequency bands with respective signal spectral values;
Step c) comprises updating each signal level estimator of the subset of frequency bands;
Step d) comprises updating the respective frequency domain gain factors of the subset of frequency bands and maintaining the previous frequency domain gain factors of the remaining frequency bands;
13. The digital acoustic input of item 12, wherein step e) comprises converting the updated frequency domain gain factor and the maintained frequency domain gain factor into updated values of the M time-varying filter coefficients of the processing filter. How to process the signal.

(Item 14)
14. The method of processing a digital acoustic input signal according to item 13, wherein the subset of frequency bands includes only a single frequency band.

(Item 15)
Item 13 wherein different subsets of frequency bands are updated during successive sample updates of M delayed digital acoustic signal samples or during successive block updates according to a predetermined iterative band update schedule. Or a processing method of a digital sound input signal according to 14;

(Item 16)
A computer-readable data carrier comprising executable program instructions configured to, when executed, cause a signal processor to perform steps a) -g) of the method of item 12.

Claims (15)

A multi-band signal processor,
A signal input for receiving a digital acoustic input signal;
A cascade of digital allpass filters configured to receive the digital acoustic input signal and generate M delayed digital acoustic signal samples at each tapping node interposed between the digital allpass filters; ;
A signal convolution processor configured to convolve the M delayed digital acoustic signal samples with M time-varying filter coefficients of a processing filter to produce a processed digital output signal;
A frequency domain transform processor configured to convert the M delayed digital acoustic signal samples into a frequency domain representation to generate respective signal spectral values in N frequency bands;
A level estimator configured to calculate respective signal level estimators in N frequency bands based on respective signal spectral values;
A processing gain calculator configured to calculate a frequency domain gain factor for each of the N frequency bands based on the respective signal level estimator and band gain rules;
An inverse frequency domain transform processor configured to transform the N frequency domain gain factors into the M time-varying filter coefficients of the processing filter, the frequency domain transform processor comprising: A multi-band signal processor configured to provide at least two of the signal spectrum values at a rate, at least two of the frequency bands, wherein M is a positive integer and N is a positive integer.   The multiband signal of claim 1, wherein the signal convolution processor is configured to be updated for each sample, or to be updated for a block, wherein each block includes a plurality of digital audio signal samples. Processor. The frequency domain transform processor comprises:
Calculating one of the signal spectrum values for a first frequency band of the N frequency bands at a first band update rate;
Configured to calculate another one of the signal spectral values for a second frequency band of the N frequency bands at an update rate lower than the first band update rate;
The multi-band signal processor according to claim 1, wherein a center frequency of the first frequency band is higher than a center frequency of the second frequency band. The signal convolution processor is configured to be updated in multiple updates, and in each of the updates,
The frequency domain transform processor is configured to update the subset of signal spectral values for the subset of the N frequency bands;
The level estimator is configured to update a subset of the signal level estimator for the subset of the N frequency bands;
The processing gain calculator updates the subset of frequency domain gain factors for the subset of N frequency bands and maintains the remainder of the frequency domain gain factors for the remainder of the N frequency bands. The multi-band signal processor according to claim 2, which is configured as follows.   The multiband signal processor of claim 4, wherein the subset of frequency bands is formed by a single frequency band of the N frequency bands. The inverse frequency domain transform processor comprises:
Configured to convert the updated frequency domain gain factor and the maintained frequency domain gain factor into the M time-varying filter coefficients by performing a set of scalar-vector multiplications;
The scalar included in the scalar-vector multiplication includes the updated frequency domain gain factor or the maintained frequency domain gain factor, and the vector included in the scalar-vector multiplication is a coefficient of a synthesis matrix based on IFFT. The multi-band signal processor of claim 4, comprising one row or column.   The multiband signal processor of claim 1, wherein the frequency domain transform processor is configured to update the signal spectrum value at a constant update rate for the respective frequency bands.   The multiband signal processor of claim 1, wherein the frequency domain transform processor is configured to update the signal spectrum value according to a predetermined iterative band update schedule.   The frequency domain transform processor is configured to provide a window between the M delayed digital acoustic signal samples and a windowed or un-windowed discrete Fourier transform coefficient in a row of the discrete Fourier transform matrix. The multiband signal processor of claim 1, wherein the multiband signal processor is configured to calculate at least one of the signal spectral values as a vector dot product of.   One or more of the band gain rules are configured to provide multiband dynamic range compression of the digital sound input signal, multiband dynamic range expansion of the digital sound input signal, or noise reduction of the digital sound input signal The multiband signal processor of claim 1, wherein: A hearing aid used by a user,
A multi-band signal processor according to claim 1;
A first microphone coupled to the multiband signal processor;
A hearing aid device comprising a speaker coupled to the multiband signal processor. A method of processing a digital acoustic input signal to produce a processed digital output signal comprising:
Allpass filtering the digital acoustic input signal through a cascade of digital allpass filters to generate M delayed digital acoustic signal samples;
Converting the M delayed digital acoustic signal samples to a frequency domain representation in N frequency bands to calculate respective signal spectral values;
Obtaining respective signal level estimators in the N frequency bands based on the signal spectrum values;
Calculating respective frequency domain gain factors for the N frequency bands based on the respective signal level estimators and band gain rules;
Converting the frequency domain gain factor into a time domain representation to generate M time-varying filter coefficients of a processing filter;
Convolving the M delayed digital acoustic signal samples with the M time-varying filter coefficients of the processing filter to generate a processed digital output signal;
Updating said M delayed digital acoustic signal samples according to a rate of each sample or a predetermined block rate, wherein at least two of said signal spectral values for at least two of said N frequency bands Is updated at different rates, M is a positive integer and N is also a positive integer. Updating the subset of signal spectral values for the subset of N frequency bands;
Updating the subset of signal level estimators for the subset of N frequency bands;
Updating the subset of frequency domain gain factors for the subset of N frequency bands;
13. The method of claim 12, further comprising: maintaining a remainder of the frequency domain gain factor for the remainder of the N frequency bands.   The method of claim 12, wherein the M delayed digital acoustic signal samples are updated according to a predetermined repetitive band update schedule.   13. A computer product comprising a non-transitory medium storing executable program instructions, wherein the method of claim 12 is performed by executing the executable program instructions by a signal processor.