CN110708651B

CN110708651B - Hearing aid squeal detection and suppression method and device based on segmented trapped wave

Info

Publication number: CN110708651B
Application number: CN201910913542.0A
Authority: CN
Inventors: 黄翔东; 高月; 陈霏; 石东宇
Original assignee: Tianjin University
Current assignee: Shenzhen Eartech Co ltd
Priority date: 2019-09-25
Filing date: 2019-09-25
Publication date: 2021-05-07
Anticipated expiration: 2039-09-25
Also published as: CN110708651A

Abstract

The invention discloses a hearing aid howling detection and suppression method and a device based on segmented trapped wave, wherein the method comprises the following steps: rounding the quotient of the product of a preset integer and a howling frequency estimation value and the sampling rate to be used as an integer parameter, and using the difference between the quotient and the integer parameter as a decimal parameter; fast calculating a coefficient vector of the FIR trap with the length of 2N-1 according to the convolution window with the length of N, integer parameters and decimal parameters; taking N-1 samples at the tail of the previous sub-segment to form a front continuation sub-segment, taking N-1 samples at the beginning of the next sub-segment to form a rear continuation sub-segment, and piecing the front continuation sub-segment, the sub-segment samples and the rear continuation sub-segment to form an extended segment with the length of P + 2N-2; and (3) convolving the coefficient vector of the wave trap with the expansion segment to generate data with the length of P +4N-4, and taking out the P data in the middle as the output of the wave trap. The device comprises: the device comprises an analog-to-digital converter, a DSP chip, an output drive module and a display module thereof.

Description

Hearing aid squeal detection and suppression method and device based on segmented trapped wave

Technical Field

The invention relates to the technical field of digital signal processing, in particular to a howling detection and suppression method and device for a hearing aid based on segmented notch, and particularly relates to a problem of how to detect and suppress the howling phenomenon of a voice signal of the hearing aid.

Background

Auditory sense is one of the important senses of human beings, is an important link for communication with the surroundings, and is not less important than visual sense. Global noise pollution and the aging of the world population structure have led to an increasing population of global hearing loss in recent years. Scientists are continually adopting various methods to help hearing impaired patients provide hearing, and wearing hearing aids is one of the most common methods to compensate for hearing loss. Nowadays, digital hearing aids gradually replace traditional analog hearing aids, but digital hearing aids also have a troublesome problem, the howling phenomenon. The basic function of a digital hearing aid is to perform dynamic range expansion corresponding to different frequencies of a speech signal, that is, to perform unequal amplification of different frequency bands in the speech signal, and the setting of these unequal gain coefficients is determined according to the hearing impairment state of a hearing impaired person during the fitting of the hearing aid. However, for the open ear hearing aids that are popular at the present time, not only does it amplify the input signal, but there is also a feedback loop from the speaker to the microphone; when amplification and feedback occur in one system simultaneously, a positive feedback phenomenon may occur, which may lead to instability of the system, reflected in the digital hearing aid as howling. When howling occurs, the signal output by the loudspeaker generates high-intensity shock phenomenon at a certain frequency tone, so that the wearer feels uncomfortable. This greatly affects the performance of the hearing aid.

There are many howling suppression methods, and the methods mainly applied to hearing aids at present mainly include class 3^[1]: gain attenuation wave trap^[2]Adaptive filter^[3-6]。

The main idea of the gain attenuation method is to reduce the gain of the echo appearing channel. The amplitude of the gain reduction is determined by the amplitude of the echo signal, which has the advantage of low power consumption and the disadvantage of reducing the gain of the desired signal, which affects the hearing compensation of the patient.

The trap method has the function principle that whether howling exists in a voice signal is detected according to a howling detection algorithm, if the howling exists, a trap of the frequency point is generated according to the howling frequency point, and the howling of the frequency point is filtered separately. However, the existing trap wave scheme has the defects that the existing trap wave scheme cannot meet the real-time processing requirement of howling detection and inhibition, which is shown in that 1) in order to ensure the real-time performance, samples for analysis cannot occupy overlong time width, and the rapid design of a trap wave device is very difficult to realize at short time intervals, and the traditional trap wave device design method cannot meet the real-time requirement on the design efficiency; 2) in order to realize the suppression of the hearing aid howling in real time, the trap point of the trap must be completely located at the howling frequency point, the decibel number of attenuation at the trap point must be large enough, and in order to ensure that the voice signal is not distorted, the pass band of the trap frequency response curve must be flat enough, which puts very strict requirements on the performance of the trap.

The main principle of the adaptive filter method is to adjust the parameters of the digital filter to obtain a frequency response close to the feedback channel, and then subtract the output of the digital filter from the output of the microphone to cancel the echo. However, when an adaptive filtering method is used for simulating an expected notch frequency response characteristic, the difficulty is high, and the defect that the attenuation decibel of a notch point is insufficient exists.

Reference to the literature

[1]Chung K.Challenges and recent developments in hearing Aids:PartⅡ.Feedback and occlusion effect reduction strategies.Laser Shell Manufacturing Processes,and Other Signal Processing Technologies Trends in Amplification,2004,8(4):125-164.

[2] From the term "Schchen 22531, Yankeepin, Guapine, Wangdonguan, Schroengzhi. research and implementation of DSP-based Howling suppression systems [ J ] Acoustic techniques, 2015,34(04): 338-.

[3]Ngo K,Van Waterschoot T,Graesboll Christensen M,Moonen M,Holdt Jensen S.Improved prediction error filters for adaptive feedback cancellation in hearing aids.Signal processing,2013,93(11):3062-3075.

[4]Ma G,Gran F,Jocobsen F,Agerkvist F.Adaptive feedback cancellation with band-limited LPC vocoder in digital hearing aids.IEEE Transactions on Audio,Speech,and Language Processing,201119(4):677-687.

[5]Guo M,Jensen J H,Jensen J.Novel acoustic feedback cancellation approaches in hearing aid applications using probe noise and probe noise enhancement.IEEE Transactions on Audio,Speech,and Language Processing,2012,20(9):2549-2563.

[6]Falco Strasser,Henning Puder.Adaptive feedback cancellation for realistic hearing aid applications[J].IEEE/ACM Transactions on Audio,Speech and Language Processing(TASLP),2015,23(12).

[7] Yellow Wedng full phase digital signal processing [ D ] Tianjin, institute of Electrical Automation and information engineering, 2006, Tianjin university.

Disclosure of Invention

The invention provides a hearing aid howling detection and suppression method and a device based on segmented trapped wave, wherein each voice segment is processed by a low-complexity algorithm and used for detecting whether howling exists or not, when the howling is detected, the howling center frequency is quickly determined, and the design of a trap with controllable center frequency points is quickly realized in a short segmented interval, which is described in detail as follows:

in a first aspect, a hearing aid howling detection and suppression method based on a segmented notch includes:

rounding the quotient of the product of a preset integer and a howling frequency estimation value and the sampling rate to be used as an integer parameter, and using the difference between the quotient and the integer parameter as a decimal parameter;

fast calculating a coefficient vector of the FIR trap with the length of 2N-1 according to the convolution window with the length of N, integer parameters and decimal parameters;

taking N-1 samples at the tail of the previous sub-segment to form a front continuation sub-segment, taking N-1 samples at the beginning of the next sub-segment to form a rear continuation sub-segment, and piecing the front continuation sub-segment, the sub-segment samples and the rear continuation sub-segment to form an extended segment with the length of P + 2N-2;

and (3) convolving the coefficient vector of the wave trap with the expansion segment to generate data with the length of P +4N-4, and taking out the P data in the middle as the output of the wave trap.

Wherein the method further comprises:

and calculating the energy of each sub-segment sample, comparing the energy with a given threshold, performing FFT with the point number of P on the sub-segment sample when the energy is greater than the threshold, and taking the frequency corresponding to the spectral peak as a howling frequency estimation value.

Further, the method further comprises:

when less than the threshold, directly outputting P samples of the sub-segment samples.

Wherein the method further comprises:

and collecting P samples of the sub-segment samples directly output or P data in the middle, splicing all the sub-segments, and outputting the sub-segments as the voice signals after the howling is restrained.

In a second aspect, a hearing aid howling detection and suppression device based on segmented notch comprises an analog-to-digital converter, a DSP chip, and an output driving and displaying module,

sampling the collected voice signals by an analog-to-digital converter to obtain a sample sequence, and entering a DSP chip in a serial digital input mode;

the DSP chip implements the method steps of the first aspect when executing a program, and finally displays the filtered speech signal by means of the output driver and its display module.

The technical scheme provided by the invention has the beneficial effects that:

1. the howling suppression algorithm has high efficiency and low calculation complexity, and avoids heavy calculation; the voice quality is improved without influencing the hearing compensation effect while the howling is completely inhibited;

2. the whole method does not depend on any additional analog circuit, directly samples voice signals, then groups the voice signals to solve energy, and judges whether howling is generated or not and also judges the howling center frequency;

3. the trap filter is very simple in design, the howling center frequency is directly substituted into a mathematical analysis formula, filtering can be realized, and a complex iterative optimization process is not needed;

4. the center frequency of the wave trap is controllable, and howling signals can be accurately and efficiently filtered.

Drawings

Fig. 1 is a flow chart of howling detection and suppression design proposed by the present invention;

fig. 2 is a schematic diagram of a hearing aid system model;

FIG. 3 is a diagram of a hearing aid model with howling detection and notch;

FIG. 4 is a signal energy profile;

figure 5 is a graph of trap frequency transmission;

FIG. 6 is a diagram of the time domain and frequency domain comparison of the original signal and the signal containing howling;

figure 7 is a waveform diagram of the trap output;

FIG. 8 is a diagram of a hardware implementation;

fig. 9 is a DSP internal flow chart.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in further detail below.

According to the howling suppression method, howling detection is carried out by calculating the energy of the voice signal within a certain time, and after priori information of the center frequency position of the signal is obtained through detection, a trap filter is designed by using a deduced analytic formula to carry out howling suppression. The wave trap is a narrow-band filter with a point-pass-like transmission characteristic, the central frequency point of which can be accurately controlled, and then the filter is used for filtering a sampling sequence. The invention makes up the defect that the common wave trap generates frequency hopping, not only realizes the real-time detection of howling, but also realizes the real-time inhibition of the howling, hardly causes the damage of the quality of useful signals, and obtains better hearing-aid effect.

Example 1

The invention proposes to perform processing according to the following steps, namely, the voice signal squeaking suppression can be realized.

Step 1: sample segmentation: with f_sSpeech is sampled at a sampling rate of 16kHz to obtain L samples which are then processed in segments to generate sub-segments x_i＝[x_i(0),x_i(1),...,x_i(P-1)]＝[x(iP),x(iP+1),...,x(iP+P-1)]，

Where the number of samples per segment P is 2000 (i.e. occupies P/f)_s125ms 1/8 seconds).

Step 2: and (3) howling detection: calculating the energy of each segment

And compared with a given threshold value T if E_iIf T is less than or equal to T, directly outputting the original segment x_iThe P samples are diverted to step 7; otherwise, if E is satisfied_i>And T, entering the step 3.

And step 3: estimation of howling frequency: a pair of subsections x_iPerforming FFT with point number of P, and taking frequency corresponding to spectral peak as howling frequency estimation value

And 4, step 4: designing a wave trap: given an integer N, determining an integer parameter

Further determining a decimal parameter

Given a convolution window { w } formed by convolution of a symmetric window of length N and a flipped rectangular window of length N_c(N), N ═ N + 1.., 0.., N-1}, where C ═ w_c(0) The parameters are substituted into the following analytical formula, and an FIR trap coefficient vector g with the length of 2N-1 is rapidly calculated as [ g (-N +1) ], g (0) ], g (N-1) ]]，

Wherein:

and 5: and (3) segmental data continuation:

taking N-1 samples at the end of the previous sub-segment to form a pre-continuation sub-segment x^f _i＝[x_i-1(P-N-3),...,x_i-1(P-2),x_i-1(P-1)]，

Taking N-1 samples from the beginning of the next sub-segment to form a post-continuation sub-segment x^b _i＝[x_i+1(0),x_i+1(1),...,x_i+1(N-2)]，

Then x is converted into^f _i、x_i、x^b _iPieced together to form an extended segment x with length P +2N-2_i'＝[x^f _i x_i x^b _i]。

Step 6: trapped wave and data selection: segmenting the filter trap coefficient vector g and the expansion segment x_iPerforming convolution to generate data with the length of P +4N-4, further removing the 2N-2 data at the front and the 2N-2 data at the tail, taking out the P data in the middle as the output of a wave trap, and entering step 7.

And 7: sub-segment splicing: and collecting the data output in the step 2 or the step 6, splicing the sub-segments, and outputting the spliced sub-segments as the voice signal after the howling is suppressed.

Example 2

The scheme in example 1 is further described below with reference to specific examples and calculation formulas, which are described in detail below:

2.1 Howling generating principle

A schematic diagram of a hearing aid system model is shown in fig. 2. The overall system can be seen as a loop system, where x (t) represents the speech signal input to the hearing aid, y (t) represents the output signal of the hearing aid, and H (e)^jω) A system function representing algorithms for compensation and noise reduction in hearing aids, G (e)^jω) Representing the true echo feedback path.

As can be seen from the model diagram of the hearing aid system, the loop system function of the hearing aid system is

At a certain frequency, howling may occur when the hearing aid system is in positive feedback. According to the Nyquist stability criterion, the following two conditions are met when the system generates howling:

1) open loop gain condition

|G(e^jω)H(e^jω)|≥1 (1)

2) Phase delay condition

∠G(e^jω)H(e^jω)＝n·360,n＝1,2,... (2)

Since the phase delay of the feedback signal varies greatly with the frequency, the condition of equation (2) is easily satisfied, and in practice, the echo path of the hearing aid is not constant, and the transfer function is affected by various factors:

1) characteristics of the user, such as jaw movements, pinna shape, size, ear canal resonance, etc.;

2) physical properties of the hearing aid itself, such as the type of hearing aid, the size of the vent, the length of the sound guide tube, etc.;

3) changes in acoustic environment, such as talking, chewing, yawning, hugging or placing a cell phone near the ear with a hearing aid, etc.;

4) hearing aid or two-mode failures such as leaks in the sound guide tube, eardrum cracks, abnormal contact of internal components, etc.

Therefore, the gain condition is also satisfied in some cases, and howling is finally formed.

2.2 Howling detection

As shown in fig. 3, which is a diagram of a hearing aid model with howling detection and notch, when the howling detection algorithm detects that howling occurs, a trap is automatically generated and inserted into the forward path to achieve the purpose of howling suppression.

According to the principle that self-oscillation with larger amplitude is generated when howling occurs, the invention judges whether the howling occurs or not by detecting the energy of each frequency band. The specific method comprises the following steps: grouping the voice signals with the total length L, and calculating the energy E of each group_iIf the frequency exceeds a certain preset value T, the frequency band is judged to have howling, then the voice signal of the frequency band is subjected to Fourier transform, and the frequency corresponding to the peak value of the frequency spectrum is the howling center frequency f₀。

As shown in fig. 4, (a) is the waveform diagram of the original voice signal, (b) is the waveform diagram of the voice signal containing howling, and (c) is the energy distribution diagram of the two signals, the energy value rises suddenly after the howling is added, and the howling is judged to be generated when the energy value exceeds the threshold value.

2.3 Howling suppression

In a digital hearing aid, after howling is detected by a howling detection algorithm, howling suppression needs to be performed under the condition of minimum influence on a signal spectrum, a wave trap is a special stop band filter meeting the conditions, and ideally, even single-frequency elimination can be performed, and the invention uses the following analytic expression:

the designed wave trap is simple and convenient to calculate and high in efficiency, and the center frequency of the trapped wave can be accurately controlled according to the m and lambda values. The transmission curve diagram of the trap frequency designed by the invention is shown in figure 5.

The analytic formula (3) of the wave trap solves the design problem of the FIR wave trap with the accurately controllable central frequency point. Assuming that the filter order is N (the final FIR filter length is 2N-1), w in equation (3)_c(n) is a single-window full-phase convolution window formed by performing inverse convolution on the front window f and the rear window b, namely w_c(n)＝b(n)*f(-n),n∈[-N,N](ii) a Taking the central element C of the convolution window as w_c(0) Is w_cNormalization factor of (n)^[7](ii) a m isAn integer representing a coarse location of the center frequency point; λ is the exact position of the center frequency point. The value of m and lambda is only needed to substitute the analytic formula shown in the formula (3), and the digital center frequency f can be designed₀A wave trap.

In order to ensure that the output of the wave trap is the same as the input, each group of sampling signals needs to be extended forward by N-1 points and extended backward by N-1 points, and the extended data segments with the length of P +2N-2 are generated by combining the extended data segments with the current subsegments:

wherein, x is_iThe method comprises the steps of performing convolution with a wave trap coefficient vector g with the length of 2N < -1 > to generate a data section with the length of P +4N < -4 >, respectively removing the first 2N < -2 > data and the last 2N < -2 > data from the data section, and taking the residual middle P data as a final trapped wave output signal of the section.

And further splicing the trapped wave output signals of the segments to obtain the final howling-removing signal.

Example 3

In order to verify the performance of the hearing aid howling detection and suppression method based on the segmented notch, a simulation experiment is carried out. The speech signal used for the experiment was an audio signal having a duration of about 7.65 s. Audio sampling frequency of f_s16kHz, total sampling length L122000, each set of sampling points P2000, adding howling at t 1.5s until the end of the speech signal, the frequency of the howling being 2kHz and the amplitude being 1, and the time domain diagram and the frequency domain diagram of the original speech signal and the speech signal containing the howling being shown in fig. 6.

After howling detection, as can be seen from fig. 6, a howling signal with a frequency of 2kHz appears at t ═ 1.5 s. The result of processing the above-mentioned method by using the segmentation notch-based de-howling method of the present invention is shown in fig. 7.

As can be seen from FIG. 7, the dots are 3.0X 10⁴Where (t ═ 3 × 10)⁴/f_s＝3×10⁴/(16×10³) 1.875 seconds ═ 1.875 seconds)

After the howling suppression is completed, the howling is suppressed after the time difference Δ t is 1.875-1.5 is 0.375 is 3 × 0.125 seconds, that is, the time difference is 3 subsegments. The existence of this time interval is due to:

(1) the threshold value of the segmented energy is set according to an empirical value, and the empirical value can only be roughly taken, so that the optimal value is difficult to obtain; (2) howling continues for a transition time from occurrence to when the energy of the sub-segment, although gradually increasing, is still below a specified threshold.

Nevertheless, the time interval of 0.375 second corresponding to 3 segments is still within the acceptable range, and the howling energy in the time interval is not large, so that the auditory effect of the human is not affected.

As can be seen from a comparison between fig. 7(a) and fig. 7(c), after Δ t, the original speech signal is almost completely restored by the signal filtered by the present invention, so the present invention method realizes howling detection and howling suppression in a parallel manner, and the howling detection and howling suppression can be realized in real time because the computation complexity consumed by the design of the wave trap and the filtering process of the sub-segment is very low.

Example 4

As shown in fig. 8, a hardware implementation diagram is that a collected voice signal x (t) is sampled by an a/D (analog-to-digital converter) to obtain a sample sequence x (n), and the sample sequence x (n) enters a DSP chip in a serial digital input form and is processed by an internal algorithm of the DSP chip to obtain a filtering signal passing through a wave trap; and finally, displaying the filtered voice signal by means of the output drive and the display module thereof.

The DSP (Digital Signal Processor) in fig. 8 is a core device, and in the Signal howling detection and suppression process, the following main functions are completed:

(1) calling a core algorithm to complete howling detection, trap filter design and howling suppression;

(2) center frequency f obtained by squeal detection₀The wave is transmitted into a wave trap to enable the wave trap to accurately filter;

(3) outputting the result to a driving and display module;

the internal program flow of the DSP device is shown in fig. 9. The howling detection and suppression of the voice signal with high precision, low complexity and high efficiency is completed based on the core estimation algorithm of the howling detection and suppression of the hearing aid based on the segmented notch, which is provided by the invention, implanted into the DSP device.

The flow of fig. 9 is divided into the following steps:

1) firstly, the sampling frequency f of the signal is set according to the specific application requirements_s；

2) Secondly, reading the sampled data from the I/O port by the CPU main controller, and entering an internal RAM;

3) finally, howling detection and suppression are performed according to the aforementioned 7-step processing procedure of the present invention, and the recovery signal is displayed through an external display device.

In the embodiment of the present invention, except for the specific description of the model of each device, the model of other devices is not limited, as long as the device can perform the above functions.

Those skilled in the art will appreciate that the drawings are only schematic illustrations of preferred embodiments, and the above-described embodiments of the present invention are merely provided for description and do not represent the merits of the embodiments.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A hearing aid howling detection and suppression method based on segmented notch is characterized by comprising the following steps:

convolving the coefficient vector of the wave trap with the expansion segment to generate data with the length of P +4N-4, and taking out the P data in the middle as the output of the wave trap;

the taking out of the middle P data as the output of the wave trap specifically comprises the following steps:

calculating the energy of each sub-segment sample, comparing the energy with a given threshold, performing FFT with the point number of P on the sub-segment sample when the energy is greater than the threshold, and taking the frequency corresponding to a spectral peak as a howling frequency estimation value; when the value is less than the threshold value, directly outputting P samples of the sub-segmentation samples; collecting P samples of the sub-segmentation samples directly output or P data in the middle, splicing all the sub-segmentations, and outputting the sub-segmentations as a voice signal after squeal suppression; p is the number of points.

2. A hearing aid howling detection and suppression device based on segmented notch comprises an analog-to-digital converter, a DSP chip, an output drive and display module, and is characterized in that,

when the DSP chip executes the program, the method steps according to claim 1 are implemented, and finally the filtered speech signal is displayed by means of the output driver and its display module.