Fusion sound signal noise reduction processing method and device for hearing aids

Abstract

The present invention discloses a fusion-type sound signal noise reduction processing method and device for hearing aids. The Bayesian filtering technology is used to fuse the amplitude spectra of the output signals of two hearing aid sound signal noise reduction processing models to obtain a fused new signal amplitude spectrum. The new signal amplitude spectrum is divided by the original signal amplitude spectrum to obtain an amplitude gain ratio. The gain ratio is used for inverse Fourier transform to obtain the noise-reduced sound signal. This method can not only eliminate non-steady-state noise, but also eliminate noise in untrained scenarios, so as to improve the sound quality and adaptability of hearing aid products.

Description

Technical Field

[0001] The present invention relates to the technical field of hearing aids, and in particular to a fusion-type sound signal noise reduction processing method and device for hearing aids. Background Art

[0002] To ensure high-quality perception of effective sound information from the outside world, existing hearing aids typically perform noise reduction processing on external sounds to improve sound clarity. For example, a general noise reduction data processing method is designed based on the assumption that "noise is stable." However, traditional noise reduction methods are generally ineffective in eliminating non-stationary noise and are prone to residual musical noise. After scene training, neural network noise reduction can effectively eliminate both steady-state and non-stationary noise, but it relies on scene training and the noise reduction effect is still poor in untrained scenes.

[0003] Bayesian filtering is a data processing method that uses the system's state equation to make optimal estimates based on the system's multi-channel input data.

[0004] Based on the above shortcomings of existing hearing aid noise reduction technology, in order to achieve better noise reduction effects, new noise reduction processing methods are urgently needed. Bayesian filtering can be used to fuse multiple noise reduction methods to process and output the results. Summary of the Invention

[0005] The purpose of the present invention is to provide a fusion sound signal noise reduction processing method and device for hearing aids, which can not only eliminate non-steady-state noise that conventional noise reduction cannot eliminate, but also eliminate noise in scenarios where artificial intelligence noise reduction has not been trained, so as to improve the sound quality and adaptability of hearing aid products.

[0006] To achieve the above objectives, the present invention provides a fusion-type sound signal noise reduction processing method for a hearing aid, comprising:

[0007] Obtaining an original amplitude spectrum of a noisy digital signal, and performing Fourier transform on the noisy digital signal to obtain a frequency domain spectrum, wherein the noisy digital signal is obtained by encoding the collected noisy sound signal;

[0008] Inputting the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and acquiring the amplitude spectrum of the output signal as the first signal amplitude spectrum; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method;

[0009] The noisy digital signal is simultaneously input into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and the amplitude spectrum of the output signal is obtained as the second signal amplitude spectrum; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method;

[0010] Using Bayesian filtering on the first signal amplitude spectrum and the second signal amplitude spectrum to obtain a fused new signal amplitude spectrum;

[0011] Dividing the new signal amplitude spectrum by the original signal amplitude spectrum to obtain the amplitude gain ratio;

[0012] The original frequency domain spectrum is multiplied by the gain ratio, and the result of the multiplication is subjected to inverse Fourier transform to obtain the noise-reduced sound signal.

[0013] In addition, to achieve the above objectives, the present invention also provides a fusion-type sound signal noise reduction device for a hearing aid, which includes the following modules:

[0014] The input module 100 is configured to obtain an original amplitude spectrum of a noisy digital signal and perform a Fourier transform on the noisy digital signal to obtain a frequency domain spectrum, wherein the noisy digital signal is obtained by encoding a collected noisy sound signal;

[0015] The fusion module 200 is used to input the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of its output signal as a first signal amplitude spectrum; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method; simultaneously input the noisy digital signal into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of its output signal as a second signal amplitude spectrum; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method; and use Bayesian filtering to obtain a fused new signal amplitude spectrum from the first signal amplitude spectrum and the second signal amplitude spectrum;

[0016] The output module 300 is used to divide the amplitude spectrum of the new signal by the amplitude spectrum of the original signal to obtain an amplitude gain ratio, and perform an inverse Fourier transform on the result obtained by multiplying the original frequency domain spectrum by the gain ratio to obtain a noise-reduced sound signal.

[0017] In addition, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to:

[0018] Obtain the original amplitude spectrum of the noisy digital signal, and perform Fourier transform on the noisy digital signal to obtain the frequency domain spectrum, wherein the noisy digital signal is obtained by encoding the collected noisy sound signal

[0019] Inputting the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and acquiring the amplitude spectrum of the output signal as the first signal amplitude spectrum; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method;

[0020] The noisy digital signal is simultaneously input into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and the amplitude spectrum of the output signal is obtained as the second signal amplitude spectrum; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method;

[0021] Using Bayesian filtering on the first signal amplitude spectrum and the second signal amplitude spectrum to obtain a fused new signal amplitude spectrum;

[0022] Dividing the new signal amplitude spectrum by the original signal amplitude spectrum to obtain the amplitude gain ratio;

[0023] The original frequency domain spectrum is multiplied by the gain ratio, and the result of the multiplication is subjected to inverse Fourier transform to obtain the noise-reduced sound signal.

[0024] The fusion sound signal noise reduction method and device for hearing aids provided by the present invention, through Bayesian filtering technology, fuses the output results of different noise reduction data processing models, and has very important application value. It can not only eliminate non-steady-state noise that conventional noise reduction cannot eliminate cleanly, but also eliminate noise in scenarios where artificial intelligence noise reduction has not been trained, thereby improving the sound quality and adaptability of hearing aid products. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the flow of the fusion sound signal noise reduction method for hearing aids according to the present invention;

[0026] Figure 2 Schematic diagram of the functional modules of the fusion sound signal noise reduction device for hearing aids according to the present invention;

[0027] The realization of the objectives, functional features and advantages of the present invention will be further explained with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0028] The present invention will be further described below in detail through specific embodiments in conjunction with the accompanying drawings. The specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0029] Figure 1 This is a flow chart of a fusion sound signal noise reduction method for a hearing aid according to an embodiment of the present invention, comprising the following steps:

[0030] S100, obtaining an original amplitude spectrum Mag_out of a noisy digital signal X_in, and performing a Fourier transform on the noisy digital signal to obtain a frequency domain spectrum Y_fq, wherein the noisy digital signal X_in is obtained by encoding a collected noisy sound signal;

[0031] S200: Input the noisy digital signal X_in to a first noise reduction data processing model in the current acoustic environment to obtain a model output signal Y_tr, and acquire the amplitude spectrum of the output signal as a first signal amplitude spectrum Mag_tr; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method;

[0032] S300: Inputting the noisy digital signal X_in simultaneously into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal Y_nn, and acquiring the amplitude spectrum of the output signal as a second signal amplitude spectrum Mag_nn; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method;

[0033] S400, using Bayesian filtering to obtain a fused new signal amplitude spectrum Mag_bayes by applying Bayesian filtering to the first signal amplitude spectrum Mag_tr and the second signal amplitude spectrum Mag_nn;

[0034] S500, dividing the new signal amplitude spectrum Mag_bayes by the original signal amplitude spectrum Mag_out to obtain an amplitude gain ratio G;

[0035] S600 , multiplying the original frequency domain spectrum Y_fq by the gain ratio G, and performing an inverse Fourier transform on the result of the multiplication to obtain a noise-reduced sound signal Y_out.

[0036] For a two-dimensional continuous random variable, the Bayesian estimation result is

[0037]

[0038] in:

[0039] f X (x) is called the prior probability density.

[0040] f Y|X (y|x) is called the likelihood probability density.

[0041] f X|Y (x|y) is called the posterior probability density.

[0042] When y is a constant, is a constant, called the Bayesian normalization constant. Therefore, the Bayesian formula for a two-dimensional continuous random variable can be expressed as:

[0043] Posterior probability = η × likelihood × prior probability

[0044] If the likelihood probability density function is set to a Gaussian function:

[0045]

[0046] At this time, the mean μ of the likelihood probability density function represents the true value of the state quantity, and σ represents the data error.

[0047] Assume that the values ​​of the first signal amplitude spectrum Mag_tr and the second signal amplitude spectrum Mag_nn are μ1 and μ2, respectively, and satisfy the normal distribution. In engineering practice, the errors of the two outputs are detected to be σ1 and σ2, respectively. According to the normal distribution, the Bayesian formula for the first and second noise reduction data processing model signals is expressed as:

[0048]

[0049]

[0050] According to Bayesian estimation, the posterior probability, which is the result after filtering, is expressed as follows:

[0051]

[0052] Where η is expressed as:

[0053]

[0054] After derivation and calculation, the final Bayesian filtering result is:

[0055]

[0056] f x|y (x|y) satisfies the normal distribution, which is expressed as:

[0057]

[0058] Therefore, the final calculation result of Mag_bayes obtained after the amplitude spectrum Mag_tr and Mag_nn are fused through Bayesian filtering is:

[0059]

[0060] Corresponding to the above method embodiment, the present invention also provides an embodiment of a fusion type sound signal noise reduction device for a hearing aid, see Figure 2 As shown, the device includes the following modules:

[0061] The input module 100 is configured to obtain an original amplitude spectrum of a noisy digital signal and perform a Fourier transform on the noisy digital signal to obtain a frequency domain spectrum, wherein the noisy digital signal is obtained by encoding a collected noisy sound signal;

[0062] The fusion module 200 is used to input the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of its output signal as a first signal amplitude spectrum; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method; simultaneously input the noisy digital signal into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of its output signal as a second signal amplitude spectrum; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method; and use Bayesian filtering to obtain a fused new signal amplitude spectrum from the first signal amplitude spectrum and the second signal amplitude spectrum;

[0063] The output module 300 is used to divide the amplitude spectrum of the new signal by the amplitude spectrum of the original signal to obtain an amplitude gain ratio, and perform an inverse Fourier transform on the result obtained by multiplying the original frequency domain spectrum by the gain ratio to obtain a noise-reduced sound signal.

[0064] In addition, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores one or more programs, and the one or more programs can also be executed by one or more processors to:

[0065] Obtain the original amplitude spectrum of the noisy digital signal, and perform Fourier transform on the noisy digital signal to obtain the frequency domain spectrum, wherein the noisy digital signal is obtained by encoding the collected noisy sound signal

[0066] Inputting the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and acquiring the amplitude spectrum of the output signal as the first signal amplitude spectrum; preferably, the first noise reduction data processing model selects a preset traditional noise reduction data processing method, that is, a non-neural network noise reduction data processing method;

[0067] The noisy digital signal is simultaneously input into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and the amplitude spectrum of the output signal is obtained as the second signal amplitude spectrum; preferably, the second noise reduction data processing model selects a preset neural network noise reduction data processing method;

[0068] Using Bayesian filtering on the first signal amplitude spectrum and the second signal amplitude spectrum to obtain a fused new signal amplitude spectrum;

[0069] Dividing the new signal amplitude spectrum by the original signal amplitude spectrum to obtain the amplitude gain ratio;

[0070] The original frequency domain spectrum is multiplied by the gain ratio, and the result of the multiplication is subjected to inverse Fourier transform to obtain the noise-reduced sound signal.

[0071] The above description is only a preferred embodiment of the present invention and does not limit the scope of the invention. All equivalent structural changes made by using the contents of the present invention and the drawings, or directly or indirectly applied to other related technical fields under the inventive concept of the present invention are included in the scope of the patent protection of the present invention. The above description is only a preferred embodiment of the present invention and does not limit the scope of the invention. All equivalent structural or equivalent process changes made by using the contents of the present invention and the drawings, or directly or indirectly applied to other related technical fields, are also included in the scope of the patent protection of the present invention.

Claims

1. A fusion-based sound signal noise reduction processing method for hearing aids based on Bayesian filtering, characterized in that: The following steps are involved: Obtaining an original signal amplitude spectrum of a noisy digital signal, and performing Fourier transform on the noisy digital signal to obtain a frequency domain spectrum, wherein the noisy digital signal is obtained by encoding the collected noisy sound signal; Inputting the noisy digital signal into a first noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtaining the amplitude spectrum of the output signal as a first signal amplitude spectrum; Inputting the noisy digital signal into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtaining an amplitude spectrum of the output signal as a second signal amplitude spectrum; Using Bayesian filtering on the first signal amplitude spectrum and the second signal amplitude spectrum to obtain a fused new signal amplitude spectrum; Dividing the new signal amplitude spectrum by the original signal amplitude spectrum to obtain the amplitude gain ratio; The original frequency domain spectrum is multiplied by the gain ratio, and the result of the multiplication is subjected to inverse Fourier transform to obtain the noise-reduced sound signal.

2. A Bayesian filtering-based fusion sound signal noise reduction processing method for a hearing aid according to claim 1, characterized in that: The first noise reduction data processing model selects a preset traditional noise reduction data processing method.

3. A Bayesian filtering-based fusion sound signal noise reduction processing method for a hearing aid according to any one of claims 1-2, characterized in that: The second denoising data processing model selects a preset neural network denoising data processing method.

4. A Bayesian filtering-based fusion sound signal noise reduction processing method for a hearing aid according to claim 1, characterized in that: The values ​​of the first signal amplitude spectrum and the second signal amplitude spectrum are μ1 and μ2 respectively, and they satisfy the normal distribution. At the same time, the errors of the two signal amplitude spectrum outputs are detected as σ1 and σ2 respectively. The final calculation result Mag_bayes of the new signal amplitude spectrum obtained after Bayesian filtering fusion is:

5. A fusion-type sound signal noise reduction device for a hearing aid, characterized in that: Includes the following modules: An input module is configured to obtain an original signal amplitude spectrum of a noisy digital signal and perform a Fourier transform on the noisy digital signal to obtain a frequency domain spectrum, wherein the noisy digital signal is obtained by encoding a collected noisy sound signal; A fusion module is configured to input the noisy digital signal into a first noise reduction data processing model in a current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of the output signal as a first signal amplitude spectrum; simultaneously input the noisy digital signal into a second noise reduction data processing model in the current acoustic environment to obtain a model output signal, and obtain the amplitude spectrum of the output signal as a second signal amplitude spectrum; and use Bayesian filtering to obtain a fused new signal amplitude spectrum from the first signal amplitude spectrum and the second signal amplitude spectrum; The output module is used to divide the amplitude spectrum of the new signal by the amplitude spectrum of the original signal to obtain the amplitude gain ratio, and perform inverse Fourier transform on the result obtained by multiplying the original frequency domain spectrum by the gain ratio to obtain the noise-reduced sound signal.

Images