Methods for processing audio signals

Abstract

The invention relates to a method (100) for processing audio signals, comprising the following method steps - Providing (110) an audio signal x(n); - Providing (120) a filter function; - applying (130) the filter function to the audio signal x(n) in the time domain to provide a filtered audio signal x̃(n); and - outputting (140) the filtered audio signal x̃(n); where - the provision of the filter function is dependent on K individual filter functions, where K ≥ 2; and - wherein the individual filter functions comprise at least two high-pass filter functions or at least two low-pass filter functions from which a band-pass filter function is determined.

Description

[0001] The present invention relates to a method for processing audio signals and a corresponding device.

[0002] Various audio signal processing methods are known from the state of the art for different application scenarios. These methods are used in particular to suppress background noise in the field of communications technology and in hearing aids. For this purpose, unfiltered audio signals are often recorded and then converted to the frequency domain before the audio signal is filtered in the frequency domain. The short-time Fourier transform (STFT) is typically used to convert the recorded time signal to the frequency domain. Signal processing in the frequency domain has the advantage that specific frequencies can be specifically suppressed or amplified.For example, an audio signal can be attenuated in the frequency ranges where noise is expected, while the signal is amplified or at least remains unattenuated in the frequency ranges where speech information is expected.

[0003] From the document DE 10 2022 111 300 A1, a device for noise reduction in headphones or hearing aids is known in which sensor signals are preprocessed, weighted by frequency band and combined to form a compensation signal for noise and occlusion suppression as well as for an ambient mode.

[0004] US Pat. No. 6,141,672 A discloses a digital filter arrangement with an adjustable cutoff frequency that can be linearly adjusted via a single parameter. The usable frequency range depends on the sampling rate and register width and also includes low frequencies. Depending on the selection of the arithmetic operator, a low-pass or high-pass characteristic is obtained.

[0005] A disadvantage of audio signal processing in the frequency domain is that audio data must be processed frame by frame, resulting in higher latencies. In addition, the audio signal (recorded as a time-dependent signal) must first be transformed into the frequency domain, filtered in the frequency domain, and then transformed back into the time domain, which increases computational complexity. The resulting latency can lead to an unnatural perception or even disorientation for a user, as the perceived acoustic signals are delayed compared to the visual perception (for example, in the case of a conversation between two people, one of whom wears a hearing aid). The superposition of passive and delayed, actively reproduced sound can result in comb filter effects that distort the frequency spectrum.In addition, if the processing latency is too high, the system cannot react quickly enough to impulse noises to protect hearing from damage, for example.

[0006] Many available audio codecs feature dedicated processors for recording, processing, and playing back audio signals with low latency. Signal conversion and processing are often performed per sampling point, so that latency can be in the low microsecond range at correspondingly high sampling rates. However, such processors typically have only rudimentary components and limited resources from which to construct an audio signal processing topology. Therefore, for more complex audio applications, such as noise suppression, the available methods typically rely on proven frameworks based on the aforementioned short-time Fourier transform. An alternative to these are filter bank approaches. The filter banks are designed, for example, based on linear-phase FIR filters.However, the approaches known from the state of the art lead to a significant latency, which can often amount to several milliseconds, which is usually undesirable in practice.

[0007] Based on the above-described disadvantages of the methods known from the prior art, it is the object of the present invention to provide a method for processing audio signals which enables efficient audio signal processing with low latency.

[0008] To achieve the above-mentioned object, the present invention proposes a method for processing audio signals, which comprises the following method steps: - Providing an audio signal x(n); - Providing a filter function; - applying the filter function to the audio signal x(n) in the time domain to provide a filtered audio signal x̃(n); and - Output the filtered audio signal x̃(n); where - the provision of the filter function is dependent on K individual filter functions, where K ≥ 2; and - wherein the individual filter functions comprise at least two high-pass filter functions or at least two low-pass filter functions from which a band-pass filter function is determined.

[0009] Unlike most methods known from the prior art, the method according to the invention filters the audio signal in the time domain. Furthermore, the method according to the invention enables efficient provision of the filter function using at least two high-pass filters or at least two low-pass filters. As will be explained below, different implementation variants can be used within the scope of the present invention to provide the filter function, depending on K individual filter functions. Even if the present invention can be described in its general form using two individual filter functions, it is clear that more than two individual filter functions can also be used.In particular, it can be provided that four, eight, or 16 individual filter functions (high-pass filters or low-pass filters) are used to provide multiple bandpass filter functions. Based on the bandpass filter functions, the filter function (also referred to as the overall filter function) can then be calculated. As explained below, the individual bandpass filter functions can be used to provide a bandpass-filtered audio signal, which is then multiplied by individual weighting functions, depending on the specific application scenario (e.g., to suppress specific noises, such as street noise, cafeteria noise, or channel noise).

[0010] In the method according to the invention, it can preferably be provided that the high-pass filter functions are provided online, i.e., in real time. This is made possible by the particularly efficient implementation of the method according to the invention and, in particular, by the computationally efficient provision of the bandpass filter function from at least two high-pass filter functions or at least two low-pass filter functions.

[0011] In the method according to the invention, it is provided that the provision of the filter function comprises a subtraction of a first individual filter function from a second individual filter function, wherein the first individual filter function and the second individual filter function are each designed as a filter function of a high-pass filter or a filter function of a low-pass filter.

[0012] The subtraction of the individual filter functions can be performed either directly or implicitly by introducing modified weighting factors, as explained in more detail below. This allows a bandpass filter function to be provided with relatively little computational effort by subtracting two high-pass filter functions or two low-pass filter functions from each other.

[0013] In some embodiments of the method according to the invention, it may be provided that - two high-pass filters (in particular two adjacent high-pass filters) are used, whose high-pass filter transfer function H k (z) is designed so that its magnitude response above a first cut-off frequency f o,k a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters (in particular two adjacent low-pass filters) are used, whose low-pass filter transfer function L k (z) is designed so that its magnitude response is below a second cut-off frequency f u,k has a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%.

[0014] In other words, two high-pass filters can be used, designed so that their magnitude responses approach each other at high frequencies, or two low-pass filters can be used, designed so that their magnitude responses approach each other at low frequencies. The first cutoff frequency and the second cutoff frequency are also referred to as stopband frequencies in the context of the present invention. The deviation of the magnitude responses (in percent) for two high-pass filters or two low-pass filters can be defined as follows: DMagnitude=|H1|−|H2||H1|×100, where D Magnitude denotes the deviation of the magnitude responses, |H1| the magnitude response of a first individual filter (high-pass filter or low-pass filter), and |H2| the magnitude response of a second individual filter (high-pass filter or low-pass filter) above or below the corresponding stopband frequency. The deviation D Magnitude The magnitude responses above or below the corresponding stopband frequency are, in the preferred embodiments, always lower than the aforementioned 10%, 5%, 3% or 1%.

[0015] By using individual filter functions whose magnitude responses approach each other at least on one side, destructive interference is achieved, allowing the bandpass filters to be deployed efficiently. This will be explained in more detail below in connection with the figures.

[0016] The high-pass filter function and the low-pass filter function are generally specified in the z-domain.

[0017] Alternatively or in addition to what has been described above, it may be provided in some embodiments of the method according to the invention that - two high-pass filters are used, whose high-pass filter transfer function H k (z) is designed so that its phase response above a first cut-off frequency f o,k a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters are used, whose low-pass filter transfer function L k (z) is designed so that its phase response is below a second cut-off frequency f u,khas a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%.

[0018] In other words, two high-pass filters can be used, designed so that their phase responses approach each other at high frequencies, or two low-pass filters can be used, designed so that their phase responses approach each other at low frequencies. The deviation of the phase responses (in percent) for two high-pass filters or two low-pass filters can be defined analogously to the deviation of the magnitude responses described above: DPhase=φ1−φ2φ1×100, where D Phasedenotes the deviation of the phase responses, φ1 denotes the phase response of a first individual filter (high-pass filter or low-pass filter), and φ2 denotes the phase response of a second individual filter (high-pass filter or low-pass filter) above or below the corresponding stopband frequency. The deviation D Phase The phase responses above or below the corresponding stopband frequency are always lower than the mentioned 10%, 5%, 3% or 1% in the preferred embodiments.

[0019] By using individual filter functions whose phase responses approach each other at least on one side, as well as by subtraction, destructive interference is achieved, thus providing the desired bandpass filters. Destructive interference through a modified phase function, in particular through a phase shift of 180°, and addition is also possible and is also considered subtraction within the meaning of the present invention.

[0020] Furthermore, in the method according to the invention, it can be provided that the individual filters are designed as finite impulse response, FIR, filters or as infinite impulse response, IIR, filters.

[0021] The advantage of using FIR filters is that they are easy to design and implement. Furthermore, FIR filters are always stable because they do not use feedback loops.

[0022] The use of IIR filters has the advantage of high efficiency. Furthermore, they allow for sharper cutoff frequencies compared to FIR filters.

[0023] Preferably, the method according to the invention can provide for the individual filters to have a non-linear phase response. Initial studies have shown that a further reduction in latency can be achieved by using individual filters with a non-linear phase response.

[0024] In some preferred embodiments of the method according to the invention, it can be provided that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kgk(n)⋅xk(n), where g k (n) denotes a weighting factor and x k (n) is determined as follows: xk(n)={yk(n)−yk+1(n), 1≤k <Kyk(n), k=K, where y x(n) each one by means of a high-pass filter H k filtered audio signal x(n), and the high-pass filters H k each with a cutoff frequency f h,k where f h,k < f h,k+1 The implementation of the filter function described above allows for efficient processing of the audio signals using high-pass filters, thus enabling low latency.

[0025] According to further preferred embodiments of the method according to the invention, it can be provided that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kgk(n)⋅xk(n), where g k (n) denotes a weighting factor and x k (n) is determined as follows: xk(n)={yk(n), k=1yk(n)−yk−1(n), 1 <k≤K, where y k (n) each one by means of a low-pass filter L kfiltered audio signal x(n), and where the low-pass filters L k each with a cutoff frequency f t,k have, in which f t,k < f t,k+1 The implementation of the filter function described above allows for efficient processing of the audio signals using low-pass filters, thus enabling low latency.

[0026] As already explained above, the subtraction of the individual filter functions can also be achieved implicitly using modified weighting factors. For this purpose, it is preferable to determine the filtered audio signal x̃(n) as follows: x˜(n)=∑k=1Kg˜k(n)⋅yk(n), where y x (n) each one by means of a high-pass filter H k filtered audio signal x(n), where the high-pass filters H k each with a cutoff frequency f h,k have, in which f h,k < f h,k+1applies, and g̃ k (n) denotes a modified gain factor, which is determined as follows: g˜k(n)={gk(n), k=1gk(n)−gk−1(n), 1 <k≤K .

[0027] This allows a particularly efficient implementation of the filter function using high-pass filters, thus achieving low latency.

[0028] It can also preferably be provided that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kg˜k(n)⋅yk(n), where y k (n) each one by means of a low-pass filter L k filtered audio signal x(n), and where the low-pass filters L k each with a cutoff frequency f t,k have, in which f t,k < f t,k+1 applies, and g̃ k (n) denotes a modified gain factor, which is determined as follows: g˜k(n)={gk(n)−gk+1(n),1≤k <Kgk(n),k=K.

[0029] This allows for a particularly efficient implementation of the filter function using low-pass filters, enabling low latency.

[0030] In some embodiments of the method according to the invention, it can be provided that at least one individual filter function is designed as a delta function (also referred to as a Dirac function or unit impulse function). In this case, K ≥ 3, K ≥ 4, K ≥ 8 or K ≥ 16 individual filter functions can preferably be provided, of which at least one individual filter function is designed as a delta function. This corresponds to a short circuit of a high-pass filter or a low-pass filter. As a result, the overall latency caused by the high-pass filters or the low-pass filters can be reduced if necessary. At the same time, the complexity of the filter is reduced if necessary. If high-pass filters are used, the first individual filter function (k = 1) can preferably be designed as a delta function. If low-pass filters are used, the last filter function (k = K) can preferably be designed as a delta function.

[0031] According to some embodiments of the method according to the invention, it can be provided that - the multiplication of the weighting factors g k (n) with the audio signals y filtered by the high-pass filters or the low-pass filters k (n) calculated signals x k (n) or the multiplication of the modified weighting factors g̃ k (n) with the audio signals filtered by the high-pass filters or the low-pass filters y k (n) with a first frequency f1; - the summation of the products of the weighting factors g k (n) with the filtered audio signals y k (n) calculated signals x k (n) or the summation of the products of the modified weighting factors g̃ k (n) with the audio signals filtered by high-pass filters or low-pass filters y k (n) with the first frequency f1; and - the calculation of the weighting factors g k (n) or the modified weighting factors g̃ k (n) with a second frequency f2; where - the second frequency f2 is lower than the first frequency f1.

[0032] In initial studies, it was shown that the second frequency (f2) or frequency with which the weighting factors or the modified weighting factors are calculated can be chosen to be lower than the first frequency (f1) or the frequency with which the multiplication of the (modified) weighting factors with the filtered audio signals y k (n) and the summation of the products of the (modified) weighting factors with the filtered audio signals y k(n) without significantly compromising the quality of the filtered audio signals. This allows for more efficient processing of the audio signals without noticeably compromising the quality of the filtered audio signals. As a result, a further reduction in latency can be achieved.

[0033] For example, it can be provided that the first frequency f1 = 192 kHz, while the second frequency f2 = 16 kHz. It can also be provided that the first frequency f1 = 96 kHz and the second frequency f2 = 8 kHz. Depending on the application scenario, the first frequency and the second frequency can be modified. Advantageously, for efficient sampling rate conversion, the ratio r fbetween f1 and f2 is an integer. The ratio can be selected according to requirements, advantageously with a value between 2 and 64, preferably with a value between 4 and 32, particularly preferably with a value between 8 and 16. In particular, as shown by the above examples, the ratio r f = 12. The second frequency f2 can preferably be ≤ 48 kHz, ≤ 24 kHz, or ≤ 16 kHz. Initial studies have shown that by selecting the above-mentioned parameter values, efficient processing of the audio signals can be achieved while simultaneously ensuring good quality of the filtered audio signals.

[0034] Furthermore, the method according to the invention can provide that the application of the filter function to the audio signal x(n) in the time domain comprises the application of a convolution operation to the audio signal x(n) and the impulse response h(n) of a high-pass filter or a low-pass filter.

[0035] Furthermore, to solve the problem described above, a device for processing audio signals is proposed, comprising: - a recording unit for recording an audio signal x(n); - a computing unit for calculating a filter function and for processing the audio signal x(n); wherein the computing unit is designed to apply the filter function to the audio signal x(n) in the time domain and to calculate a filtered audio signal x̃(n); - an output unit for outputting the filtered audio signal x̃(n); characterized in that - the computing unit is designed to determine the filter function as a function of K individual filter functions, where K ≥ 2; and - to determine a bandpass filter function from at least two individual filter functions comprising two high-pass filter functions or two low-pass filter functions.

[0036] The device according to the invention allows particularly efficient processing of audio signals with low latency.

[0037] Preferably, in the device according to the invention, it can be provided that the computing unit is designed to calculate a bandpass filter function by subtracting a first individual filter function from a second individual filter function, wherein the first individual filter function and the second individual filter function are each designed as a filter function of a high-pass filter or a filter function of a low-pass filter.

[0038] In addition, the device according to the invention can be provided with the properties described above in connection with the method according to the invention and with the computing unit of the device according to the invention being designed to carry out the method steps described in connection with the method.

[0039] In the method according to the invention, it can be provided in particular that the high-pass filters or the low-pass filters used are designed as non-linear phase filters.

[0040] Preferably, in the present invention, it can be provided that the high-pass filters or the low-pass filters are designed as recursive, minimum-phase filters.

[0041] It can also preferably be provided that the first high-pass filter (k = 1) or the last low-pass filter (k = K) is replaced by a short circuit. In this way, the computational complexity can be reduced, thereby achieving lower latency.

[0042] Furthermore, it can preferably be provided that the weighting factors or the modified weighting factors are calculated as a function of at least one sensor signal, in particular as a function of a microphone signal. This allows a response to a dynamic interference signal and the filter function can be adapted to the recorded interference signal.

[0043] The present invention is explained in more detail below with reference to the figures. Fig. 1 is a flowchart for an embodiment of the present invention, Fig. 2 a signal flow diagram for an embodiment of the present invention, Fig. 3 shows an exemplary magnitude response of high-pass filters according to an embodiment of the present invention, Fig. 4 shows an exemplary phase response of high-pass filters according to an embodiment of the present invention, Fig. 5 shows an exemplary magnitude response of resulting bandpass filters according to an embodiment of the present invention, Fig. 6 a signal flow diagram for an embodiment of the present invention based on low-pass filters, Fig. 7 is a signal flow diagram for an embodiment of the present invention with modified weighting factors, Fig. 8 is a signal flow diagram for an embodiment of the present invention, wherein the first high-pass filter is replaced by a short circuit, Fig. 9 a schematic representation of an embodiment of the device according to the invention, Fig. 10 a schematic representation of a further embodiment of the device according to the invention comprising several sensors, preprocessing units and a loudspeaker, and Fig. 11 a schematic representation of a further embodiment of the device according to the invention in the form of an in-ear headphone comprising several sensors, a loudspeaker and a computing unit for calculating and applying the filter function.

[0044] Fig. 1 shows a flowchart for an embodiment of the method 100 according to the invention. In a first method step 110, an audio signal is provided. This can be done using a microphone or another sensor. Alternatively, it can be provided that the audio signal is provided by an external source that has previously recorded the audio signal. In a second method step 120, a filter function is provided. The filter function is provided as a function of K individual filter functions, where K ≥ 2. The individual filter functions can be implemented as high-pass filter functions or low-pass filter functions. A bandpass filter function is calculated from the individual filter functions. This can be done in particular by subtracting two high-pass filter functions or two low-pass filter functions.In a third method step 130, the provided filter function is applied to the audio signal in the time domain. This produces a filtered audio signal. In a fourth method step 140, the filtered audio signal is output. The filtered audio signal can be output either via a loudspeaker or another output unit, or via signal transmission to an external device.

[0045] Fig. Figure 2 shows an exemplary embodiment of the method according to the invention. In this embodiment, an input signal x(n) is transmitted to several high-pass filters 10 through a parallel circuit, wherein Fig. 2 shows a first high-pass filter 10a, a second high-pass filter 10b, and a K-th high-pass filter 10c. H k with k ∈ [1..K] has the cutoff frequencies f h,k where f h,k < f h,k+1The specific design of the cutoff frequencies can vary depending on the application scenario. At the output of the high-pass filters 10, the high-pass filtered signals y k (n). Neighboring high-pass filtered signals are then subtracted from each other to produce bandpass signals xk(n)={yk(n)−yk+1(n),1≤k <Kyk(n),k=K The bandpass signals are then weighted with time-varying weighting factors g k (n) weighted and then added to the output signal x˜(n)=∑k=1Kgk(n)⋅xk(n) summed up. The weighting factors g k(n) can preferably assume a value from 0 to 1. In some embodiments of the present invention, it can also be provided that the weighting factors can also assume values ​​that are greater than 1 or less than 0. The weighting factors are also calculated depending on the specific application. When calculating the weighting factors, the approaches known from the prior art can be used. In order to generate corresponding bandpass signals from the individual high-pass filters by subtraction, it is preferred that the individual high-pass filters have certain magnitude and phase properties. These preferred properties are exemplary in the Fig. 3 and Fig. 4. In particular, the Fig. 3 shows that the magnitude responses of the individual high-pass filters approach 20 at high frequencies. Fig. 3 shows a first magnitude range 20a, a second magnitude range 20b, a third magnitude range 20c and a fourth magnitude range 20d. As shown in the Fig. As can be seen in Figure 3, the magnitude responses 20 above a certain frequency (in particular above an upper stopband frequency f o,k ) are a short distance apart.

[0046] In addition, the Fig. 4 shows that the phase responses 20 also approach each other at high frequencies. Fig. Figure 4 shows a first phase response 21a, a second phase response 21b, a third phase response 21c, and a fourth phase response 21d. The bandpass transfer function can be defined as the difference between two high-pass filter transfer functions in the z-range as follows: Bk(z)={Hk(z)−Hk+1(z), 1≤k <KHk(z), k=K.

[0047] Due to the high-pass characteristic of the high-pass filter, the band-pass filter also has a high-pass characteristic with a corresponding lower stopband frequency f u,k = f h,k . The adjacent high-pass filters can in particular be designed so that their magnitude and / or phase response is above an upper stopband frequency f o,k , so that the subtraction leads to destructive interference. In summary, the k-th bandpass filter has a passband of f u,k ≤ f ≤ f o,k . Due to the definition of the bandpass filters by the highpass filters, the lower stop frequency of the following bandpass filter corresponds to the upper stop frequency of the bandpass filter according to f u,k+1 = f o,k .

[0048] The cutoff frequencies of the high-pass filters, and correspondingly the passbands of the bandpass filters, can advantageously be evenly distributed on a psychoacoustically motivated frequency scale, such as the Bark scale. This allows the input signal to be processed in frequency bands that mimic the human ear. However, the choice of cutoff frequencies depends on the specific application scenario. The present invention is not limited to a specific choice of cutoff frequencies.

[0049] A decisive advantage of the topology according to the invention is that the overall transfer function for unity gain g k = 1 simplifies to B¯(z)=∑k=1KBk(z) =(∑k=1K−1Hk(z)−Hk+1(z))+HK(z) =H1(z)−H2(z)+H2(z)−​ ⋯−HK(z)+HK(z) =H1(z).

[0050] The overall transfer function is therefore defined exclusively by the first high-pass filter. This results in a smooth overall transfer function in the passband and a low group delay.

[0051] The high-pass filters can be designed, for example, using an optimization method with a cost function based on the magnitude response of the bandpass filters in the passband and stopband, as well as the sum of all bandpass filters. The filters can be implemented as FIR or IIR filters. They can advantageously be minimum-phase, exhibiting a nonlinear phase response. A filter can be described as minimum-phase if its zeros, i.e., the zeros of the numerator polynomial of its filter transfer function, lie within the unit circle or have an amplitude ≤ 1. This definition applies to FIR filters as well as to IIR filters that are not implemented as all-pole filters, i.e., filters whose transfer function comprises only denominator coefficients and, if applicable, a gain factor. All-pole filters are, by definition, minimum-phase.

[0052] The high-pass filters can also advantageously be optimized so that the overall transfer function B(z) follows a desired magnitude and phase response, so that the filter bank, for example, implicitly performs frequency weighting or equalization.

[0053] Fig. 5 shows the resulting magnitude responses 22 of the exemplary bandpass filters based on the Fig. 3 and Fig. 4 high-pass filters. Overall, the Fig. 5 shows a first magnitude response 22a, a second magnitude response 22b, a third magnitude response 22c, and a fourth magnitude response 22d for the corresponding bandpass filters. The overall transfer function B(z) then corresponds to the high-pass filter with a solid line from Fig. 3 and Fig. 4.

[0054] Instead of high-pass filters, low-pass filters 11 (also known as L k(z)) can be used to implement the method according to the invention, as exemplified in Fig. 6 is shown. In the Fig. 6 shows a first low-pass filter 11a, a second low-pass filter 11b and a K-th low-pass filter 11c. The basic principle of band-pass behavior based on the characteristics of the filters and destructive interference by subtraction described above still applies, but some adjustments have been made compared to the Fig. 2. In particular, the subtraction is adjusted so that Bk(z)={Lk(z),k=1Lk(z)−Lk−1(z),1 <k≤K.

[0055] For unity gain g k = 1, the overall transfer function in this case simplifies to B(z) = L K (z).

[0056] Fig. 7 shows a further embodiment of the method according to the invention, wherein, compared to the Fig. 2 illustrated embodiment, the explicit difference formation of the high-pass signals y k (n) has been removed. Instead, modified weighting factors g̃ k (n) ∈ [-1,1] is used, so that the difference is implicitly formed during weighting. For the arrangement based on high-pass filters, the modified weighting factors can be calculated as follows: g˜k(n)={gk(n),k=1gk(n)−gk−1(n),1 <k≤K.

[0057] This rule can be determined by rearranging the equation for the overall transfer function, here without loss of generality in the z-range and neglecting the time variance of the weighting factors: B¯(z)=∑k=1Kgk⋅Bk(z) =g1⋅[H1(z)−H2(z)]+g2⋅[H2(z)−H3(z)]+⋯+gK⋅HK(z) =H1(z)⋅g1+H2(z)⋅[g2−g1]+⋯+HK(z)⋅[gK−gK−1] =H1(z)⋅g˜1+H2(z)⋅g˜2+⋯+HK(z)⋅g˜K =∑k=1Kg˜k⋅Hk(z)

[0058] Fig. Figure 8 shows a further embodiment of the method according to the invention, wherein the Fig. 2 has been replaced by a short circuit. Accordingly, the last low-pass filter 11c in the arrangement can also be Fig. 6 with a short circuit. This gives the structure a transfer function B(z) = 1 for unity gain and therefore no inherent latency. It should be noted that the overall transfer function changes for different weighting factors g k ≠ 1, which can increase latency. However, even in these cases, the input-to-output latency is very low due to the filter structure.

[0059] Fig. 9 shows an embodiment of the device 50 according to the invention comprising a recording unit 30, which is designed as a microphone, a filter bank 31, an output filter 32 and an output unit 33, which is designed as a loudspeaker.

[0060] Fig. 10 shows a further embodiment of the device 50 according to the invention, which has two sensors 30, two input filters 34, a filter bank 31 and a loudspeaker 33.

[0061] These arrangements can be found, for example, in modern headphones. Microphones record ambient sound, pre- and post-processing units (also known as input and output filters) filter the microphone signals, enabling, for example, active noise cancellation to reduce the volume of ambient sound, or an ambient mode for natural communication with the environment. A filtered audio signal is then played back via the headphone's speaker. The method according to the invention can be used in such an application, for example, to remove microphone or wind noise from the microphone signal, to perform speech enhancement, or to compress the microphone signal so that the level of loud signals is reduced before playback.

[0062] Finally, Fig. 11 shows an example of an in-ear headphone 60, which is equipped with several external microphones 40, an internal microphone 41, a vibration sensor 42, a loudspeaker 33 and a computing unit 43. In the Fig.12 also shows an ear insert 44, an auditory canal 45, and an eardrum 46. The computing unit 43 is designed to carry out the method steps according to the invention. In particular, the computing unit 43 is designed to provide the filter function and apply this to an audio signal in the time domain. Various pre- and post-processing steps of the sensor signals can optionally be carried out, such as filtering, amplification, compression, or limiting. These can be implemented according to the approaches described in the prior art. A single-channel signal based on the processing by the method according to the invention is then reproduced via the loudspeaker 33. Multiple instances of the method according to the invention can also be used to process various sensor signals, which are subsequently combined and fed to an output unit.

[0063] To determine the weighting factors g k (n), any sensor signals can be used. Within the meaning of the present invention, bandpass signals can preferably be used to calculate the respective weighting factors. In particular, a weighting factor g k (n) a function of the bandpass signal x k (n). The respective weighting factor can also depend on other bandpass signals. While the filter bank, the application of the weighting factors and the summation can be processed at a first frequency, it can preferably be provided that the weighting factors are preferably calculated at a second frequency. The second frequency can preferably be lower than the first frequency. For this purpose, either the bandpass signals x k (n), the high / low pass signals y k(n), or even the input signal x(n) can be passed to a second process via a sampling rate converter. The second process can then emulate parts of the filter bank or even the entire filter bank accordingly, so that the bandpass signals at the second frequency are available.

[0064] Since the weighting factors typically change slowly, it is advantageous to set the second sampling rate lower than the first sampling rate to reduce computational complexity. However, the Nyquist frequency should be taken into account when designing the sampling rates, ensuring that the signals are transmitted with the appropriate bandwidth and no information is lost. The weighting factors calculated at the second sampling rate do not necessarily need to be adapted to the first rate by a sampling rate conversion.

[0065] For example, the weighting factors can be calculated to allow appropriate speech components to pass through each band and attenuate all background noise, such as wind noise, ambient noise, or microphone noise, thereby improving speech quality or intelligibility. Furthermore, the weighting factors can be calculated to attenuate loud signal components per band, for example, to protect the hearing of users of an ambient mode in headphones or hearing protection with a communication function.

[0066] The inventive method can also be applied to other audio sources, such as music, telephone calls, computer games, and films. The inventive method can be used to calibrate or personalize a music playback system, an ambient mode, and active noise cancellation in headphones. Furthermore, the method can be used, for example, in hearing aid applications to compensate for hearing loss. The method can also be used for the (also frequency-dependent) calibration of audio devices. LIST OF REFERENCE SYMBOLS 10 high-pass filters 10a first high-pass filter 10b second high-pass filter 10c K-th high-pass filter 11 Low-pass filter 11a first low-pass filter 11b second low-pass filter 11c K-th low-pass filter 20 Magnitude response of the high-pass filters 20a first magnitude range 20b second magnitude range 20c third magnitude range 20d fourth magnitude range 21 Phase response of the high-pass filters 21a first phase response 21b second phase response 21c third phase transition 21d fourth phase response 22 Magnitude response of the determined bandpass filters 22a first magnitude range 22b second magnitude range 22c third magnitude range 22d fourth magnitude range 30 recording unit 31 filter bank 32 output filters 33 Output unit 34 input filters 40 outer microphone 41 inner microphone 42 Vibration sensor 43 computing unit 44 Ear insert 45 ear canal 46 eardrum 50 device according to the invention 60 in-ear headphones 100 inventive method 110 first procedural step 120 second process step 130 third procedural step 140 fourth procedural step

Claims

[1] Method (100) for processing audio signals, comprising the following method steps - Providing (110) an audio signal x(n); - Providing (120) a filter function; - applying (130) the filter function to the audio signal x(n) in the time domain to provide a filtered audio signal x̃(n); and - outputting (140) the filtered audio signal x̃(n); where - the provision of the filter function is dependent on K individual filter functions, where K ≥ 2; and - wherein the individual filter functions comprise at least two high-pass filter functions or at least two low-pass filter functions from which a band-pass filter function is determined., wherein the provision (120) of the filter function comprises a subtraction of a first individual filter function from a second individual filter function, wherein the first individual filter function and the second individual filter function are each implemented as a filter function of a high-pass filter or a filter function of a low-pass filter, where - two high-pass filters are used, whose high-pass filter transfer function H k (z) is designed so that its magnitude response above a first cut-off frequency f o,k a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters are used, whose low-pass filter transfer function L k (z) is designed so that its magnitude response is below a second cut-off frequency f u,khas a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%; and / or - two high-pass filters are used, whose high-pass filter transfer function H k (z) is designed so that its phase response above a first cut-off frequency f o,k a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters are used, whose low-pass filter transfer function L k (z) is designed so that its phase response is below a second cut-off frequency f u,k has a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%. [2] Method (100) according to claim 1,characterized by that the individual filters are designed as finite impulse response, FIR, filters or as infinite impulse response, IIR, filters. [3] Method (100) according to one of claims 1 or 2, characterized by that the individual filters have a non-linear phase response. [4] Method (100) according to one of claims 1 to 3, characterized by that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kgk(n)⋅xk(n), where g k (n) denotes a weighting factor and x k (n) is determined as follows: xk(n)={yk(n)−yk+1(n),1≤k <Kyk(n),k=K, where y k (n) each one by means of a high-pass filter HP k filtered audio signal x(n), and the high-pass filters HP k each with a cutoff frequency f h,k where f h,k < f h,k+1 applies. [5] Method (100) according to one of claims 1 to 3, characterized bythat the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kgk(n)⋅xk(n), where g k (n) denotes a weighting factor and x k (n) is determined as follows: xk(n)={yk(n),k=1yk(n)−yk−1(n),1 <k≤K, where y k (n) each one by means of a low-pass filter L k filtered audio signal x(n), where the low-pass filters L k each with a cutoff frequency f t,k have, in the f t,k < f t,k+1 applies. [6] Method (100) according to one of claims 1 to 3, characterized by that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kg˜k(n)⋅yk(n), where y k (n) each one by means of a high-pass filter H k filtered audio signal x(n), where the high-pass filters H k each with a cutoff frequency f h,k have, in the f h,k < f h,k+1applies, and g̃ k (n) denotes a modified gain factor, which is determined as follows: g˜k(n)={gk(n),k=1gk(n)−gk−1(n),1 <k≤K. [7] Method (100) according to one of claims 1 to 3, characterized by that the filtered audio signal x̃(n) is determined as follows: x˜(n)=∑k=1Kg˜k(n)⋅yk(n), where y k (n) each one by means of a low-pass filter L k filtered audio signal x(n), where the low-pass filters L k each with a cutoff frequency f t,k have, in which f t,k < f t,k+1 applies, and g̃ k (n) denotes a modified gain factor, which is determined as follows: g˜k(n)={gk(n)−gk+1(n),1≤k <Kgk(n), k=K. [8] Method (100) according to one of claims 1 to 7, characterized by that at least one individual filter function is designed as a delta function. [9] Method (100) according to one of claims 5 to 8, characterized by , that - the multiplication of the weighting factors g k (n) with the audio signals y filtered by the high-pass filters or the low-pass filters k (n) calculated signals x k (n) or the multiplication of the modified weighting factors g k (n) with the audio signals filtered by high-pass filters or low-pass filters y k (n) with a first frequency f1; - the summation of the products of the weighting factors g k (n) with the filtered audio signals y k (n) calculated signals x k (n) or the summation of the products of the modified weighting factors g̃ k (n) with the audio signals filtered by high-pass filters or low-pass filters y k (n) with the first frequency f1; and - the calculation of the weighting factors g k(n) or the modified weighting factors g̃ k (n) with a second frequency f2; where - the second frequency f2 is lower than the first frequency f1. [10] Method (100) according to one of claims 1 to 9, characterized by that the application of the filter function to the audio signal x(n) in the time domain comprises the application of a convolution operation to the audio signal x(n) and the impulse response h(n) of a high-pass filter or a low-pass filter. [11] Device (50) for processing audio signals, comprising: - a recording unit (30) for recording an audio signal x(n); - a computing unit (43) for calculating a filter function and for processing the audio signal x(n); wherein the computing unit (43) is designed to apply the filter function to the audio signal x(n) in the time domain and to calculate a filtered audio signal x̃(n); - an output unit (33) for outputting the filtered audio signal x̃(n); characterized by , that - the computing unit (43) is designed to determine the filter function as a function of K individual filter functions, where K ≥ 2, - to determine a bandpass filter function from at least two individual filter functions comprising two high-pass filter functions or two low-pass filter functions, - the computing unit (43) is designed to calculate a bandpass filter function by subtracting a first individual filter function from a second individual filter function, wherein the first individual filter function and the second individual filter function are each designed as a filter function of a high-pass filter or a filter function of a low-pass filter; and wherein the device further comprises: - two high-pass filters whose high-pass filter transfer function H k(z) is designed so that its magnitude response above a first cut-off frequency f o,k a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters whose low-pass filter transfer function L k (z) is designed so that its magnitude response is below a second cut-off frequency f u,k has a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%; and / or - two high-pass filters whose high-pass filter transfer function H k (z) is designed so that its phase response above a first cut-off frequency f o,ka deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of maximum 3% or maximum 1%, or - two low-pass filters whose low-pass filter transfer function L k (z) is designed so that its phase response is below a second cut-off frequency f u,k has a deviation of maximum 10% from each other, preferably a deviation of maximum 5% and particularly preferably a deviation of 3% or maximum 1%.

Images