Device and method for binaural speech enhancement

Abstract

Provided is a hearing device (200) for: receiving a first input signal (201) and a second input signal (202) (401); determining a first ear signal (203) based on the first input signal (201) and determining a second ear signal (204) based on the second input signal (202) (402); obtaining statistical information (205) indicating a pre - evaluated SNR probability distribution that depends on a global SNR (403); estimating a current SNR (206) based on the first input signal (201) and the second input signal (202) (404); determining one or more common gain filters (207), wherein the one or more common gain filters (207) are determined based on the statistical information (205) and the estimated current SNR (206) (405); and applying the one or more common gain filters (207) to the first ear signal (203) and the second ear signal (204) (406).

Description

Technical Field

[0001] The present invention relates to the field of hearing devices, and more particularly to binaural hearing devices. In this field, the present invention provides a hearing device and a corresponding method for binaural speech enhancement. The hearing device of the present invention is used to determine a common gain filter and apply the common gain filter to the first ear signal and the second ear signal to achieve speech enhancement. Background Art

[0002] The speech pickup of a hearing (assistive) device (such as a true wireless headset (TWS) or a hearing aid) is more or less damaged by ambient noise. The method of trying to remove noise from the microphone signal of such a hearing device through a signal processing device is called speech enhancement or noise reduction. The method of synchronously enhancing speech for both ears of a user - such as in a hearing aid, an ear - worn device, or a TWS device - is called binaural processing. Generally speaking, it is necessary to significantly reduce noise while maintaining speech quality and clarity. In a binaural environment, spatial cues should also be retained, otherwise sound localization will be affected.

[0003] Traditional noise reduction methods have high noise reduction performance, including spectral subtraction, Wiener filter, and the so - called minimum mean - square error (MMSE) short - time spectral amplitude (STSA) estimator.

[0004] In principle, this process generally consists of two parts. First, a noise estimate is determined. Then, based on the signal to be compensated and the determined noise estimate, a noise reduction filter or gain is composed. Therefore, applying the noise reduction filter to the signal can achieve speech enhancement. When composing the noise reduction filter, an important intermediate step is to estimate the local time - varying prior signal - to - noise ratio (SNR).

[0005] In some cases, for example, for hearing aids or TWS devices, binaural signal noise reduction should be performed. However, independent processing of ear signals may destroy spatial cues and thus is not a satisfactory solution. One way to retain spatial cues is to process the binaural signals using the same noise reduction filter (i.e., the so - called common gain filter). On the one hand, common gain filtering of noisy speech signals under MMSE has been proven to be a feasible solution, and its form is similar to Wiener filter spectral enhancement. On the other hand, this technique especially requires the estimation of the local time - varying prior SNR.

[0006] Traditional noise reduction methods and solutions still suffer from speech distortion and processing artifacts, such as the "musical noise" phenomenon, which is a significant drawback of Wiener-type noise reduction methods. Additionally, fine-tuning the noise reduction to mitigate the musical noise problem also requires multiple trial-and-error attempts. Summary of the Invention

[0007] In view of the above challenges and limitations, embodiments of the present invention aim to introduce devices and methods for binaural speech enhancement that construct a common gain filter to improve the degree of noise reduction and speech quality. Specifically, one objective is to avoid trial-and-error adjustments and infinite a priori SNR estimation. Another objective is to preserve spatial cues and perform significant noise reduction in a binaural speech enhancement setting.

[0008] This objective is achieved by the embodiments provided in the appended independent claims. Advantageous implementations of the embodiments are further defined in the dependent claims.

[0009] A first aspect of the present invention provides a hearing device for binaural speech enhancement, the hearing device configured to: receive a first input signal and a second input signal; determine a first ear signal based on the first input signal and a second ear signal based on the second input signal; obtain statistical information indicating a pre-evaluated SNR probability distribution that depends on a global SNR; estimate a current SNR based on the first input signal and the second input signal; determine one or more common gain filters, wherein the one or more common gain filters are determined based on the statistical information and the estimated current SNR; and apply the one or more common gain filters to the first ear signal and the second ear signal.

[0010] Accordingly, embodiments of the present invention introduce a hearing device for a binaural speech enhancement setting, wherein the hearing device is configured to use statistical information of a pre-known process to determine one or more common gain filters, i.e., the hearing device is capable of leveraging a known statistical SNR probability distribution. Thus, when applying the one or more common gain filters to binaural signals, the hearing device can improve the noise reduction effect and speech quality. In embodiments of the present invention, a priori SNR estimation is replaced by Bayesian SNR marginalization while preserving spatial cues. Additionally, trial-and-error adjustments can be avoided.

[0011] In an implementation of the first aspect, the hearing device is further configured to: determine a first sound channel between the first ear portion of the hearing device and the sound source based on the first input signal and the second input signal; determine a second sound channel between the second ear portion of the hearing device and the sound source based on the first input signal and the second input signal; estimate the current SNR based on the noise density of the first input signal and the second input signal, the first sound channel and the second sound channel, and the first input signal and the second input signal.

[0012] It should be noted that the current SNR estimation can thus depend on "online data", that is, on the sound channels, the noise input signals, and the estimated noise density. The estimated noise density can be a noise spectral density, specifically a power spectral density (PSD).

[0013] In an implementation of the first aspect, the hearing device is further configured to calculate the noise density of the first input signal and the second input signal based on the first input signal and the second input signal.

[0014] Specifically, the noise density can thus be a real-time estimate.

[0015] In an implementation of the first aspect, the one or more common gain filters are binaural Bayesian filters.

[0016] In an implementation of the first aspect, the one or more common gain filters are based on Bayesian SNR marginalization.

[0017] Specifically, the prior SNR estimate can be replaced by Bayesian SNR marginalization. The hearing device can combine the constraints of SNR marginalization, cue preservation, and noise reduction into advanced spectral weighting.

[0018] In an implementation of the first aspect, the hearing device is further configured to: maintain a lookup table including one or more common gain filters, each of the one or more common gain filters being associated with a specific global SNR; use the lookup table to determine the one or more common gain filters based on the estimated current SNR.

[0019] Specifically, a lookup table can be generated and input into the hearing device. In this way, the hearing device can be used to find a suitable common gain filter based on the calculated posterior SNR and the global SNR.

[0020] In one implementation of the first aspect, the hearing device is further configured to: obtain updated statistics indicating an updated pre - evaluated SNR probability distribution, where the updated pre - evaluated SNR probability distribution depends on an updated global SNR.

[0021] It should be noted that if the global SNR changes sharply, the pre - evaluated SNR probability distribution may be variable. By obtaining updated statistics, the hearing device can maintain good noise reduction and high speech quality.

[0022] A second aspect of the present invention provides a binaural speech enhancement method performed by a hearing device. The method includes: receiving a first input signal and a second input signal; determining a first ear signal based on the first input signal and a second ear signal based on the second input signal; obtaining statistics indicating a pre - evaluated SNR probability distribution, where the pre - evaluated SNR probability distribution depends on a global SNR; estimating a current SNR based on the first input signal and the second input signal; determining one or more common gain filters, where the one or more common gain filters are determined based on the statistics and the estimated current SNR; applying the one or more common gain filters to the first ear signal and the second ear signal.

[0023] In one implementation of the second aspect, the method further includes: determining a first sound channel between a first ear portion of the hearing device and a sound source based on the first input signal and the second input signal; determining a second sound channel between a second ear portion of the hearing device and the sound source based on the first input signal and the second input signal; estimating the current SNR based on the noise density of the first input signal and the second input signal, the first sound channel and the second sound channel, the first input signal and the second input signal.

[0024] In one implementation of the second aspect, the method further includes: calculating the noise density of the first input signal and the second input signal based on the first input signal and the second input signal.

[0025] In one implementation of the second aspect, the one or more common gain filters are binaural Bayesian filters.

[0026] In one implementation of the second aspect, the one or more common gain filters are based on Bayesian SNR marginalization.

[0027] In one implementation of the second aspect, the method further includes: maintaining a lookup table including one or more common gain filters, each of the one or more common gain filters being associated with a specific global SNR; using the lookup table to determine the one or more common gain filters based on the estimated current SNR.

[0028] In one implementation of the second aspect, the method further includes: obtaining updated statistics indicating an updated pre-evaluation SNR probability distribution, the updated pre-evaluation SNR probability distribution depending on an updated global SNR.

[0029] The method according to the second aspect and its implementations achieves the same advantages and effects as the hearing device according to the first aspect and its corresponding implementations described above.

[0030] A third aspect of the present invention provides a computer program product, the computer program product including program code, the program code, when implemented on a processor, executing the method according to the second aspect and any one of the implementations of the second aspect.

[0031] A fourth aspect of the present invention provides a non-transitory storage medium, the non-transitory storage medium storing executable program code, the executable program code, when executed by a processor, executing the method according to any one of the second aspect or its implementations.

[0032] It should be noted that all devices, elements, units, and apparatuses described in this application can be implemented in software or hardware elements or any combination thereof. All steps performed by the various entities described in this application and the functions to be performed by the various entities described are intended to mean that the corresponding entities are used to perform the corresponding steps and functions. Although in the description of the following specific embodiments, the specific functions or steps performed by external entities are not reflected in the description of the specific detailed elements of the entities performing the specific steps or functions, those skilled in the art should clearly understand that these methods and functions can be implemented by the corresponding hardware or software elements or any combination thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In combination with the drawings, the above aspects and implementations of the present invention are described in the following description of specific embodiments. In the attached

[0034] FIGURES:

[0035] Figure 1 Shows a binaural signal model using a common gain filtering method.

[0036] Figure 2 Shows the device provided by the embodiment of the present invention.

[0037] Figure 3 Shows a binaural signal model using a common gain filtering method provided by an embodiment of the present invention.

[0038] Figure 4 Shows the method provided by an embodiment of the present invention. Detailed implementation

[0039] Describes illustrative embodiments of a hearing device and method for binaural speech enhancement in conjunction with the accompanying drawings. Although this description provides detailed examples of possible implementations, it should be noted that these details are for illustration purposes only and do not limit the scope of the present application.

[0040] In addition, one embodiment / example may refer to multiple other embodiments / examples. For example, any description mentioned in one embodiment / example, including but not limited to terms, elements, processes, explanations, and / or technical advantages, applies to multiple other embodiments / examples.

[0041] Figure 1 Shows an overview of a binaural signal model using a common gain filtering method, which is also executed by a hearing device (e.g., Figure 2 the hearing device 200 shown in ) provided by an embodiment of the present invention. Specifically, in Figure 1 it is assumed that the clean speech signal has a single-source origin s(n) at sampling time n. The clean speech signal arrives at the hearing device, specifically, at the left-ear part and the right-ear part of the hearing device respectively in a binaural setup. The corresponding clean signals can be expressed as: d l / r (n) = s(n) * h l / r,n , which is the result of the convolution of the single source with the head-related impulse response (HRIR) h l / r,n of the left-ear part and the right-ear part of the hearing device. However, uncorrelated noise n l / r (n) is added to the clean signal, so the noise signal y l / r (n) = d l / r (n) + n l / r (n) is picked up by the hearing device. The noise signal can be buffered and transformed into the frequency domain through short-time Fourier transformation (STFT). This process can be expressed as: Y i (k, m) = Hi(k, m)S(k, m) + N i (k, m), where i ∈ l, r. The corresponding estimated ear signal D i (k, m) = G(k, m)Y i (k, m) can be obtained by filtering using a common gain filter G(k, m) at frequency k and frame by frame m. The obtained filtered ear signal Di (k,m) can be further transformed back to the time domain by inverse short-time Fourier transform (ISTFT), and can be processed by overlap-and-add synthesis. For convenience, the exponent is removed hereinafter.

[0042] Performing binaural speech enhancement, that is, removing noise from the noise signal to obtain the desired ear signals (for the left ear and the right ear), ultimately depends on the suitability of the common gain filter (i.e., G). This may depend on the estimates or components required to calculate the common gain filter. The common gain filter will determine the quality of speech enhancement and the ease of achieving this in terms of calculation and robustness.

[0043] As mentioned above, the traditional scheme critically requires estimating the local time-varying prior SNR. To address the above limitations of the traditional scheme, embodiments of the present invention propose a new method for the common gain filter structure that utilizes the statistical information of the pre-known or determined speech and noise characteristics.

[0044] Figure 2 Fig. 200 shows a hearing device 200 for binaural speech enhancement provided by an embodiment of the present invention. The hearing device 200 may include a processing circuit (not shown) for performing, carrying out, or initiating various operations of the hearing device 200 described herein. The processing circuit may include hardware and software. The hardware may include analog circuits or digital circuits, or both analog and digital circuits. The digital circuit may include components such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), or a general-purpose processor. The hearing device 200 may also include a memory circuit that stores one or more instructions that can be executed by the processor or the processing circuit (specifically, executed under the control of software). For example, the memory circuit may include a non-transitory storage medium storing executable software code that, when executed by the processor or the processing circuit, causes the hearing device 200 to perform various operations. In one embodiment, the processing circuit includes one or more processors and a non-transitory memory connected to the one or more processors. The non-transitory memory may carry executable program code that, when executed by the one or more processors, causes the hearing device 200 to perform, carry out, or initiate the operations or methods described herein.

[0045] Specifically, the hearing device 200 is configured to receive a first input signal 201 and a second input signal 202. These input signals 201 and 202 may correspond to the Figure 1 pickup signal y l / r (n) shown in. That is to say, these input signals 201 and 202 may include speech signals with a single-source origin and may include uncorrelated noise. It should be noted that the left-ear side portion of the hearing device 200 may pick up the first input signal 201, and the right-ear side portion of the hearing device 200 may pick up the second input signal 202. The hearing device 200 is further configured to determine a first ear signal 203 (e.g., a left-ear signal) based on the first input signal 201 and a second ear signal 204 (e.g., a right-ear signal) based on the second input signal 202.

[0046] In addition, the hearing device 200 is configured to obtain statistical information 205, wherein the statistical information 205 indicates a pre-evaluated SNR probability distribution depending on the global SNR. Specifically, the pre-evaluated SNR probability distribution may be known or have been previously estimated. The statistical information 205 may be received by the hearing device 200 through configuration (e.g., through pre-configuration). The statistical information 205 may also be received by the hearing device 200 from a statistical information source and may be updated periodically.

[0047] The hearing device 200 is further configured to estimate a current SNR 206 based on the first input signal 201 and the second input signal 202, that is, the SNR 206 currently present in these input signals 201 and 202. It should be noted that the current SNR 206 may be a local time-varying posterior SNR estimate based on the first input signal 201 and the second input signal 202.

[0048] Then, the hearing device 200 is configured to determine one or more common gain filters 207, wherein the one or more common gain filters 207 are determined based on the statistical information 205 and the estimated current SNR 206. The hearing device 200 is further configured to apply the one or more common gain filters 207 to the first ear signal 203 and the second ear signal 204 in order to obtain a filtered first ear signal and a filtered second ear signal. The filtered first ear signal and the filtered second ear signal reduce noise, thereby enhancing speech.

[0049] According to an embodiment of the present invention, the design concept of the hearing device 200 is to introduce the statistical information 205 of a pre-known process, that is, to utilize the known statistical SNR probability distribution to determine the common gain. Therefore, the embodiments of the present invention propose to replace the prior SNR estimation used in traditional hearing devices and methods with posterior SNR estimation and prior SNR marginalization. It should be noted that the statistical information 205 can be an offline parameter, which is known or can be pre-estimated. Therefore, the posterior SNR estimation (i.e., the current SNR 206) can be the only variable depending on the online data (e.g., the acoustic channels and the input signals 201 and 202).

[0050] Figure 3 FIG. shows the hearing device 200 provided by an embodiment of the present invention, based on Figure 2 the embodiment shown. Specifically, Figure 3 FIG. shows a more detailed binaural signal model, which uses a common gain filtering method and is located within the hearing device 200. Figure 2 and Figure 3 the same elements in share the same reference numerals and can be implemented in a similar manner. The hearing device 200 may include a noise PSD estimator and a binaural Bayesian filtering unit.

[0051] The hearing device 200 receives the input signals 201 and 202 (y l and y r ) again, and may derive the first ear signal 203 and the second ear signal 204 (Y l and Y r ) from the input signals 201 and 202 based on performing STFT. The hearing device 200 also obtains the statistical information 205 (E{η+ dB}). The noise PSD estimator may estimate the noise density (Φ nn ). Specifically, in this case, the noise density Φ nn may be the noise PSD. The binaural Bayesian filtering unit may specifically determine the estimated current SNR 206 (γ) through posterior SNR estimation, and may further determine one or more common gain filters 207 based on the statistical information 205 and the estimated current SNR 206 (γ) specifically through SNR marginalization More details will be described below.

[0052] According to an embodiment of the present invention, one or more common gain filters 207 may be binaural Bayesian filters, specifically they may be binaural Bayesian filters using Bayesian SNR marginalization.

[0053] Optionally, a binaural Bayesian filter using SNR marginalization (i.e., one or more common gain filters 207) can be determined by the hearing device 200 (specifically, the binaural Bayesian filtering unit) as:

[0054]

[0055] The variables and functions in Equation (1) are defined as follows.

[0056] η and η dB (η dB = 10 log10(η)) are the prior SNR and its logarithmic representation, respectively. Due to the summation operation, prior SNR estimation is not required. This is the effect of marginalization, which is a statistical term in this context.

[0057] According to an embodiment of the present invention, the lognormal hyperprior of the prior SNR (i.e., the statistical information 205) can be expressed as:

[0058]

[0059] Specifically, Equation (2) is characterized by the mean value dB represented by the global SNR E{η} and the variance of the empirical estimate

[0060] It is possible that, based on the sufficient likelihood function, where Q is the number of most recent observations (e.g., typically set to 4), Equation (2) of the statistical information 205 can be supplemented with a multivariate observation model:

[0061]

[0062] It should be noted that the simplified Equation (3) is characterized by the calculated posterior SNR, i.e.,

[0063] According to an embodiment of the present invention, the hearing device 200 can be used to determine the first sound channel between the first ear portion of the hearing device 200 and the sound source based on the first input signal 201 and the second input signal 202. In addition, the hearing device 200 can be used to determine the second sound channel between the second ear portion of the hearing device 200 and the sound source based on the first input signal 201 and the second input signal 202. In addition, the hearing device 200 can be used to estimate the current SNR 206. Specifically, the estimation of the current SNR 206 can be based on the noise density (specifically, the noise spectral density) of the first input signal 201 and the second input signal 202, the first sound channel and the second sound channel, and the first input signal 201 and the second input signal 202.

[0064] In a specific embodiment, the current SNR 206 can be based on:

[0065]

[0066] Specifically, in equation (4), γ represents the current SNR 206, i.e., the calculated posterior SNR, where κ is the current frame time index and Q is the number of the most recent frames included. It can be seen that the current SNR 206 can depend on the acoustic channel H = [H l H r T and the noise density Φ nn . In this case, the noise density can be a noise PSD estimate, which can be calculated in real time by the noise PSD estimator of the hearing device 200.

[0067] According to an embodiment of the present invention, the hearing device 200 can also be used to calculate the noise density of the first input signal 201 and the second input signal 202 based on the first input signal 201 and the second input signal 202.

[0068] Therefore, based on the previous embodiments and equations, a look-up table can be pre-generated using the calculated offline data According to an embodiment of the present invention, the hearing device 200 can also be used to maintain such a look-up table, which includes one or more common gain filters 207. Specifically, each of the one or more common gain filters 207 can be associated with a specific global SNR.

[0069] It should be noted that the look-up table can include common gain filter entries. One or more common gain filters 207 (e.g., one or more binaural Bayesian filters using SNR marginalization) can depend on a specific global SNR (the dependence comes from the statistical information 205 indicating the SNR probability distribution) and the posterior SNR, i.e., the estimated current SNR 206. Therefore, the look-up table may also be related to these two types of SNR.

[0070] According to an embodiment of the present invention, the hearing device 200 can also be used to determine one or more common gain filters 207 based on the look-up table and the estimated current SNR 206. In this way, a suitable common gain filter can be found based on the calculated posterior SNR (γ). This can save a large amount of computation.

[0071] According to an embodiment of the present invention, the hearing device 200 can also be used to obtain updated statistical information 205 indicating an updated pre-evaluated SNR probability distribution, where the updated pre-evaluated SNR probability distribution depends on the updated global SNR.

[0072] ​Note that if the global SNR changes drastically, the pre - evaluated SNR probability distribution will also change. It is possible to estimate the global SNR and accordingly change the SNR probability distribution. It is possible that the SNR probability distribution can be estimated / replaced by using a histogram of data collected over time by the hearing device 200.

[0073] Figure 4 FIG. 400 shows a method 400 provided by an embodiment of the present invention. In a particular embodiment of the present invention, the method 400 can be performed by Figure 2 or Figure 3 the hearing device 200 shown. The method 400 includes: step 401, receiving a first input signal 201 and a second input signal 202; step 402, determining a first ear signal 203 based on the first input signal 201 and determining a second ear signal 204 based on the second input signal 202; step 403, obtaining statistical information 205 indicating a pre - evaluated SNR probability distribution, where the pre - evaluated SNR probability distribution depends on the global SNR; step 404, estimating a current SNR 206 based on the first input signal 201 and the second input signal 202; step 405, determining one or more common gain filters 207, where the one or more common gain filters 207 are determined based on the statistical information 205 and the estimated current SNR 206; step 406, applying the one or more common gain filters 207 to the first ear signal 203 and the second ear signal 204.

[0074] In summary, the present invention uses a common gain filtering method without trial - and - error and infinite prior SNR estimation. This is achieved by deriving a new formula for the common gain filter structure that utilizes statistical information 205 of pre - known or determined speech and noise characteristics.

[0075] Compared with other traditional binaural speech enhancement techniques, the advantage of the embodiment of the present invention lies in being independent of prior SNR estimation. The resulting scheme is easy to implement, improves the degree of noise reduction and speech quality, and does not require trial - and - error adjustment. Overall, compared with traditional schemes, the proposed scheme shows superiority in terms of the evaluated speech quality metrics confirmed in daily listening. Therefore, it is proven to be an equalized spectral weighting for binaural speech enhancement applications.

[0076] Embodiments of the present invention propose using Bayesian SNR marginalization to replace prior SNR estimation while retaining spatial cues and achieving significant noise reduction in a binaural speech enhancement setting. It combines the constraints of SNR marginalization, cue preservation, and noise reduction into an advanced spectral weighting.

[0077] The present invention has been described in connection with various embodiments and implementations taken as examples. However, other variations can be understood and implemented by those skilled in the art when practicing the claimed invention, based on a study of the drawings, the present invention, and the appended claims. In the claims as well as in the description, the word "comprising" does not exclude other elements or steps, and "a" does not exclude a plurality. A single element or other unit can fulfill the functions of several entities or items described in the claims. Listing certain measures in mutually different dependent claims does not indicate that a combination of these measures cannot be used effectively.

[0078] In addition, any method provided by an embodiment of the present invention can be implemented in a computer program having an encoding device, which, when run by a processing device, causes the processing device to execute the method steps. The computer program is included in a computer-readable medium of a computer program product. The computer-readable medium can basically include any memory, such as a read-only memory (ROM), a programmable read-only memory (PROM), an erasable PROM (EPROM), a flash memory, an electrically erasable EPROM (EEPROM), or a hard disk drive.

[0079] Furthermore, those skilled in the art recognize that embodiments of the hearing device 200 correspondingly include the necessary communication capabilities in the form of, for example, functions, devices, units, elements, etc. for executing the solution. Examples of other such devices, units, elements, and functions are: processors, memories, buffers, control logic, encoders, decoders, rate matchers, de-rate matchers, mapping units, multipliers, decision units, selection units, switches, interleavers, de-interleavers, modulators, demodulators, inputs, outputs, antennas, amplifiers, receiver units, transmitter units, DSPs, trellis-coded modulation (TCM) encoders, TCM decoders, power units, power feeds, communication interfaces, communication protocols, etc., which are arranged appropriately together to execute the solution.

[0080] Specifically, one or more processors of the hearing device 200 may include, for example, one or more instances of a central processing unit (CPU), a processing unit, processing circuitry, a processor, an application specific integrated circuit (ASIC), a microprocessor, or other processing logic that can interpret and execute instructions. The term "processor" may thus represent processing circuitry that includes multiple processing circuits, such as any, some, or all of the items listed above. The processing circuitry may also perform data processing functions for input, output, and processing of data, including data buffering and device control functions, such as call processing control, user interface control, and the like.

Claims

1. A hearing device (200) for binaural speech enhancement, characterized in that, the hearing device (200) is configured to: receive a first input signal (201) and a second input signal (202); determine a first ear signal (203) based on the first input signal (201) and determine a second ear signal (204) based on the second input signal (202); obtain statistical information (205) indicating a pre - evaluated signal - to - noise ratio (SNR) probability distribution, the pre - evaluated SNR probability distribution depending on a global SNR; estimate a current SNR (206) based on the first input signal (201) and the second input signal (202); determine one or more common gain filters (207), wherein the one or more common gain filters (207) are determined based on the statistical information (205) and the estimated current SNR (206); apply the one or more common gain filters (207) to the first ear signal (203) and the second ear signal (204).

2. The hearing device (200) according to claim 1, characterized in that, it is further configured to: determine a first sound channel between a first ear portion of the hearing device (200) and a sound source based on the first input signal (201) and the second input signal (202); determine a second sound channel between a second ear portion of the hearing device (200) and the sound source based on the first input signal (201) and the second input signal (202); estimate the current SNR (206) based on the noise density of the first input signal (201) and the second input signal (202), the first sound channel and the second sound channel, the first input signal (201) and the second input signal (202).

3. The hearing device (200) according to claim 2, characterized in that, it is further configured to: calculate the noise density of the first input signal (201) and the second input signal (202) based on the first input signal (201) and the second input signal (202).

4. The hearing device (200) according to any one of claims 1 to 3, characterized in that, the one or more common gain filters (207) are binaural Bayesian filters.

5. The hearing device (200) according to claim 4, characterized in that, the one or more common gain filters (207) are based on Bayesian SNR marginalization.

6. The hearing device (200) according to any one of claims 1 to 5, characterized in that, it is further configured to: maintain a look - up table including one or more common gain filters (207), each of the one or more common gain filters (207) being associated with a specific global SNR; use the look - up table to determine the one or more common gain filters (207) based on the estimated current SNR (206).

7. The hearing device (200) according to any one of claims 1 to 6, characterized in that, also for: obtaining updated statistics (205) indicating an updated pre - evaluated SNR probability distribution, the updated pre - evaluated SNR probability distribution depending on an updated global SNR.

8. A binaural speech enhancement method (400) performed by a hearing device (200), characterized in that, comprising: receiving (401) a first input signal (201) and a second input signal (202); determining (402) a first ear signal (203) based on the first input signal (201) and determining a second ear signal (204) based on the second input signal (202); obtaining (403) statistics (205) indicating a pre - evaluated signal - to - noise ratio (SNR) probability distribution, the pre - evaluated SNR probability distribution depending on a global SNR; estimating (404) a current SNR (206) based on the first input signal (201) and the second input signal (202); determining (405) one or more common gain filters (207), wherein the one or more common gain filters (207) are determined based on the statistics (205) and the estimated current SNR (206); applying (406) the one or more common gain filters (207) to the first ear signal (203) and the second ear signal (204).

9. The method (400) according to claim 8, characterized in that, further comprising: determining a first sound channel between a first ear part of the hearing device (200) and a sound source based on the first input signal (201) and the second input signal (202); determining a second sound channel between a second ear part of the hearing device (200) and the sound source based on the first input signal (201) and the second input signal (202); estimating the current SNR (206) based on the noise density of the first input signal (201) and the second input signal (202), the first sound channel and the second sound channel, the first input signal (201) and the second input signal (202).

10. The method (400) according to claim 9, characterized in that, further comprising: calculating the noise density of the first input signal (201) and the second input signal (202) based on the first input signal (201) and the second input signal (202).

11. The method (400) according to any one of claims 8 to 10, characterized in that, the one or more common gain filters (207) are binaural Bayesian filters.

12. The method (400) according to claim 11, characterized in that, the one or more common gain filters (207) are based on Bayesian SNR marginalization.

13. The method (400) according to any one of claims 8 to 12, characterized in that, further comprising: Maintain a look-up table including one or more common gain filters (207), each of the one or more common gain filters (207) being associated with a specific global SNR; Determine the one or more common gain filters (207) using the look-up table based on the estimated current SNR (206).

14. The method (400) according to any one of claims 8 to 13, characterized in that, further comprising: Obtain updated statistics (205) indicating an updated pre-evaluation SNR probability distribution, the updated pre-evaluation SNR probability distribution depending on an updated global SNR.

15. A computer program product, characterized in that, comprises program code for performing the method (400) according to any one of claims 8 to 14 when implemented in a processor.

Images