Method and system for suppressing broadcast howling and eliminating external noise

Abstract

The invention discloses a method and system for suppressing broadcast howling and eliminating external noise, relates to the technical field of howling and noise processing, and solves the technical problems that in the prior art, only howling frequency in audio is recognized and suppressed, noise in the audio is ignored, misrecognition of the howling frequency is caused, and follow-up audio processing work is affected. The method comprises the following steps: dividing the noise degree of a noise audio, analyzing a filter gain coefficient required by the noise audio, and eliminating external noise of broadcast to-be-input voice to obtain a noise-free audio; extracting a howling frequency point in the noise-free audio, and eliminating the howling frequency point to obtain a target broadcast voice; the noise is removed in advance, the howling frequency identification accuracy is improved, and the howling frequency is eliminated, so that the high-quality transmission of the broadcast audio is realized, and the auditory experience of a user is improved.

Description

Technical Field

[0001] The invention belongs to the field of howling and noise processing, and in particular relates to a method and a system for suppressing broadcast howling and eliminating external noise. Background Art

[0002] Howling in audio equipment is a high-frequency, sharp sound, usually caused by acoustic feedback or incompatibility between devices.

[0003] The invention patent application with publication number CN116312592A discloses a howling processing method and device for a voice broadcast sound reinforcement system, which relates to the field of howling processing technology. The method includes: obtaining a signal of a voice to be processed at t frames; generating a short-time spectrum to be processed of the voice signal to be processed at t frames; detecting whether there is a howling frequency point in the short-time spectrum to be processed at t frames; if so, extracting the spectrum of each howling frequency point and the spectrum of non-howling frequency points from the short-time spectrum to be processed at t frames, and for each howling frequency point, using the spectrum of the non-howling frequency point to adjust the spectrum amplitude of the howling frequency point to suppress the howling spectrum component at the howling frequency point, and converting the adjusted short-time spectrum into a first time domain voice signal for output; if there is no howling frequency point, the short-time spectrum to be processed is directly converted into a second time domain voice signal for direct output.

[0004] The above-mentioned prior art realizes the recognition and suppression of the howling frequency in the audio, but there will be noise in the audio, which will cover or confuse the howling frequency point, easily leading to the misidentification of the howling frequency, thereby affecting the subsequent audio processing work. Therefore, the present invention provides a method and system for suppressing broadcast howling and eliminating external noise. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method and system for suppressing broadcast howling and eliminating external noise, which is used to solve the technical problem that the prior art only identifies and suppresses the howling frequency in the audio, but ignores the existence of noise in the audio, resulting in misidentification of the howling frequency and affecting subsequent audio processing work.

[0006] To achieve the above object, a first aspect of the present invention provides a method for suppressing broadcast howling and eliminating external noise, comprising the following steps:

[0007] Step 1: Identify whether there is noise audio in the voice audio to be input for broadcast; if yes, the voice audio to be input is the audio with noise, and proceed to step 2; if no, the voice audio to be input is the audio without noise, and proceed to step 3;

[0008] Step 2: classify the noise level of the noisy audio; according to the noise level, analyze the gain coefficient corresponding to the filter, eliminate the external noise of the broadcast input voice, obtain the noise-free audio, and enter step 3; wherein the noise level includes first-level noise, second-level noise and third-level noise;

[0009] Step 3: Extract the howling frequency points in the noise-free audio, and eliminate the howling frequency points to obtain the target broadcast voice.

[0010] Preferably, the step of identifying whether there is noise audio in the voice audio to be input for broadcasting includes:

[0011] Extracting input voice audio of a number of broadcasts from historical data; wherein the input voice audio includes normal voice audio and noise audio;

[0012] The input speech audio is used as the input data of the artificial intelligence model, and the noise audio is used as the output data of the artificial intelligence model; the artificial intelligence model is trained by the input data and the output data to obtain a noise recognition model;

[0013] The speech audio to be input is input into the noise recognition model, and the noise audio in the speech audio to be input is output.

[0014] Through training with a large amount of historical data, the noise recognition model of the present invention can learn the subtle differences between noise and normal speech, thereby showing a high degree of accuracy in the recognition process; the method realizes the automatic recognition and extraction of noise, and avoids the errors that may be caused by human factors; and with the continuous advancement of technology and the continuous accumulation of data, the model can be further trained and optimized to improve its recognition performance and accuracy.

[0015] Preferably, the classification of the noise audio into noise levels includes:

[0016] Extract each sound pressure of the noise audio in the input speech audio, marked as Pi; wherein i is the number of the sound pressure, i=0, 1, ..., N, N is a positive integer;

[0017] The total sound pressure level SYJ of the noise audio is calculated by the formula SYJ=20×lg[(ΣPi) / (N×P0)]; where P0 is the reference sound pressure;

[0018] The total sound pressure level is compared with the preset range. If it is less than the preset range, it is the first-level noise; if it is within the preset range, it is the second-level noise; if it exceeds the preset range, it is the third-level noise.

[0019] The present invention extracts the sound pressures of the noise audio and calculates the total sound pressure level. The method can accurately quantify the intensity level of the noise audio. This quantification method enables us to understand the actual situation of the noise audio more accurately, providing strong support for subsequent processing; and by accurately quantifying the intensity of the noise audio and scientifically classifying the noise level, the method can significantly improve the processing efficiency. In practical applications, the appropriate processing strategy can be selected according to the level of the noise audio, thereby avoiding unnecessary waste of resources.

[0020] Preferably, analyzing the gain coefficient corresponding to the filter according to the noise level includes:

[0021] Extracting the total sound pressure level data of several different levels of noise audio from the historical data, and dividing the total sound pressure level of each level of noise audio into several sound pressure groups; wherein the total sound pressure level of each sound pressure group is averaged to obtain the average total sound pressure level of each sound pressure group;

[0022] Based on the average total sound pressure level, the gain coefficient of the filter is adjusted to obtain the gain coefficient required by the average total sound pressure level.

[0023] The present invention realizes refined processing of noise audio by extracting total sound pressure level data of noise audio of different levels and dividing the noise audio of each level into several sound pressure groups, which helps to more accurately understand the characteristics and distribution of noise audio and provides strong support for the design of filters; based on the average total sound pressure level of each group, the appropriate filter gain coefficient is adjusted to better eliminate noise, thereby improving voice quality, and the relationship between the average total sound pressure level and the filter gain coefficient is established; convenience and accuracy are provided for the subsequent determination of the filter gain coefficient according to the average total sound pressure level.

[0024] Preferably, the total sound pressure level of each level of noise audio is divided into a number of sound pressure groups, including:

[0025] The total sound pressure levels of noise audios of the same level are sorted by size, and each of the same step length is grouped together to obtain several sound pressure groups.

[0026] Preferably, the total sound pressure level of the noise audio in the speech to be input is calculated; it is determined whether the total sound pressure level is equal to the average total sound pressure level of the corresponding level; if so, the gain coefficient corresponding to the average total sound pressure level is used as the gain coefficient of the filter for the speech audio to be input; if not, the difference between the total sound pressure level and the average total sound pressure level of the corresponding level is calculated, and the gain coefficient corresponding to the average total sound pressure level with the smallest difference is extracted as the gain coefficient of the filter for the speech audio to be input.

[0027] Preferably, the step of extracting the howling frequency points in the noise-free audio includes:

[0028] Traverse the frequency points of the noise-free audio from left to right in turn to extract the howling frequency point; mark the frequency point currently being analyzed as the frequency point to be evaluated;

[0029] The power ratio of the frequency point to be evaluated is compared with the power values ​​of other frequency points in turn and the difference is processed. When a power value of another frequency point is smaller than the power value of the frequency point to be evaluated, and the difference between the two is greater than the difference threshold, the frequency point to be evaluated is a howling frequency point.

[0030] Among them, Fourier transform (such as fast Fourier transform FFT) is applied to convert the audio signal from time domain to frequency domain; in the frequency domain, the spectrum of the audio signal is analyzed to determine the amplitude and phase information of each frequency point; the power spectral density (PSD) describes the power distribution of the signal at different frequencies, and in the power spectral density diagram, the power value corresponding to the required frequency point is found.

[0031] The present invention can accurately identify the howling phenomenon in the audio by comparing the power values ​​of the frequency point to be evaluated with other frequency points and setting a difference threshold to judge the howling frequency point, avoiding the problem of inaccurate howling detection due to misjudgment in traditional methods, improving the accuracy of detection, and is suitable for processing and analysis of large-scale audio data, with strong flexibility and adaptability; by accurately identifying and processing the howling frequency points, the method helps to improve the overall quality of the audio.

[0032] Preferably, the step of eliminating the howling frequency points to obtain the target broadcast voice includes:

[0033] Ratio the amplitude of the howling frequency point to the amplitude of the normal frequency point, take lg, and then multiply by 20 to get the decibel value of the amplitude of the howling frequency point higher than the normal frequency point;

[0034] The decibel value obtained is summed with the decibel compensation value to obtain the attenuation of the filter, thereby eliminating the howling frequency point.

[0035] The present invention calculates the amplitude ratio of the howling frequency point to the normal frequency point and converts it into a decibel value, thereby accurately quantifying the frequency characteristics of the howling and the degree of difference between it and the normal audio. By increasing the decibel compensation value, the attenuation of the filter can be more accurately adjusted to make it more accurately match the actual amplitude of the howling frequency point, which helps to avoid excessive attenuation or insufficient attenuation, thereby improving the accuracy of howling elimination, providing an accurate basis for the design of the filter, ensuring that while eliminating the howling, the quality of the original audio signal is retained as much as possible, and significantly improving the user experience.

[0036] Preferably, a band-stop filter, a low-pass filter or a high-pass filter is selected according to the frequency band width corresponding to the howling frequency point.

[0037] Preferably, the second aspect of the present invention provides a system for suppressing broadcast howling and eliminating external noise, comprising a noise elimination module, and a noise audio recognition module and a howling suppression module connected thereto;

[0038] The noise audio recognition module is used to identify whether there is noise audio in the voice audio to be input for broadcasting; if yes, the voice audio to be input is the audio with noise, and the noise degree of the audio with noise is classified; if no, the voice audio to be input is the audio without noise;

[0039] The noise elimination module is used to analyze the gain coefficient corresponding to the filter according to the noise level, eliminate the external noise of the broadcast input voice, and obtain noise-free audio;

[0040] The howling suppression module is used to extract the howling frequency points in the noise-free audio and eliminate the howling frequency points to obtain the target broadcast voice.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. The present invention extracts a number of different levels of noise audio from historical data and analyzes the total sound pressure level data of the noise audio of different levels, subdivides it into a plurality of sound pressure groups according to the total sound pressure level data, calculates the average total sound pressure level of each group of noise audio as a basis for adjusting the filter gain coefficient, and can effectively eliminate noise by accurately matching the relationship between the average total sound pressure level and the filter gain coefficient, thereby significantly improving the quality of speech;

[0043] 2. The present invention can accurately locate the howling frequency point by analyzing each frequency point of the audio one by one; not only the absolute power value of the frequency point, but also the relative power difference between the frequency points is considered, so that even a weak but significant howling frequency can be accurately captured. By setting a reasonable difference threshold, it can effectively avoid misjudging normal audio signals as howling, thereby improving the accuracy of recognition;

[0044] 3. The present invention identifies noise audio in the broadcast voice audio to be input; divides the noise level of the noisy audio; according to the noise level, analyzes the gain coefficient corresponding to the filter, eliminates the external noise of the broadcast voice to be input, and obtains noise-free audio; extracts the howling frequency points in the noise-free audio, and eliminates the howling frequency points to obtain the target broadcast voice; the system provides a complete audio processing flow from noise recognition, noise elimination to howling suppression, ensuring that every link from input to output is optimized, and the presence of noise may mask or confuse the howling frequency points. Pre-removing noise can ensure that the howling detection module is more likely to identify the real howling frequency and eliminate it, thereby achieving high-quality transmission of broadcast audio and improving the user's auditory experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 It is a schematic diagram of the process of the present invention;

[0047] Figure 2 It is a schematic diagram of the flow of the noise audio elimination method of the present invention;

[0048] Figure 3 The figure is a flow chart of a method for eliminating howling frequencies of audio signals according to the present invention. DETAILED DESCRIPTION

[0049] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 The first aspect of the present invention provides a method for suppressing broadcast howling and eliminating external noise, comprising the following steps:

[0051] Step 1: Identify whether there is noise audio in the voice audio to be input for broadcast; if yes, the voice audio to be input is the audio with noise, and proceed to step 2; if no, the voice audio to be input is the audio without noise, and proceed to step 3;

[0052] See also Figure 2 Specifically, a plurality of broadcast input voice audios are extracted from the historical data; wherein the input voice audios include normal voice audios and noise audios;

[0053] The input speech audio is used as the input data of the artificial intelligence model, and the noise audio is used as the output data of the artificial intelligence model; the artificial intelligence model is trained by the input data and the output data to obtain a noise recognition model;

[0054] The speech audio to be input is input into the noise recognition model, and the noise audio in the speech audio to be input is output.

[0055] For example: mark the noise audio in the input voice audio of the broadcast, such as the equipment sound in the broadcasting room, the sound of cars outside, the sound of horns, etc.; mark the noise audio as 0 and the normal audio as 1; input several marked input voice audios of the broadcast into the artificial intelligence model, the artificial intelligence model learns the characteristics of the noise audio and the normal audio, and outputs the voice audio marked as 0, that is, outputs the noise audio; obtain the noise recognition model through repeated training; among them, the artificial intelligence model can select the RNN neural network model.

[0056] Step 2: classify the noise level of the noisy audio; wherein the noise level includes first-level noise, second-level noise and third-level noise;

[0057] Specifically, each sound pressure of the noise audio in the input speech audio is extracted, marked as Pi; wherein i is the number of the sound pressure, i=0, 1, ..., N, N is a positive integer;

[0058] The total sound pressure level SYJ of the noise audio is calculated by the formula SYJ = 20 × lg [(ΣPi) / (N × P0)]; where P0 is the reference sound pressure, usually 20 × 10 -6 Pa, this is the pressure of the weakest sound that the human ear can detect at 1000Hz.

[0059] The total sound pressure level is compared with the preset range. If it is less than the preset range, it is the first-level noise; if it is within the preset range, it is the second-level noise; if it exceeds the preset range, it is the third-level noise.

[0060] For example: suppose there are three different noises in the noise audio of a broadcast voice audio, which are marked as noise A, noise B and noise C, and there are several sound pressures in each noise. For example, the sound pressure of noise A fluctuates, so there are several different sound pressures, and these different sound pressures are extracted; the same is true for noise B and noise C; suppose there are 3 different sound pressures in noise A, 2 sound pressures in noise B, and 2 sound pressures in noise C, so there are 7 different sound pressures in the noise audio, and the 7 sound pressures are averaged, and the sound pressures are numbered, respectively P0, P1, P2, P3, P4, P5, P6, and the sound pressures are averaged to obtain the average sound pressure of the noise audio = (P0+P1+P2+P3+P4+P5+P6) / 7;

[0061] The total sound pressure level of the three noise sound intensities is calculated using the formula SYJ=SYJ=20×lg (average sound pressure / P0).

[0062] According to the noise level, the gain coefficient corresponding to the filter is analyzed to eliminate the external noise of the broadcasted voice to be input, obtain the noise-free audio, and proceed to step three;

[0063] Specifically, the total sound pressure level data of several noise audios of different levels are extracted from the historical data, the total sound pressure levels of the noise audios of the same level are sorted by size, and each of the same step lengths is grouped into a group to obtain several sound pressure groups;

[0064] The total sound pressure level of each sound pressure group is averaged to obtain the average total sound pressure level of each sound pressure group;

[0065] Based on the average total sound pressure level, the gain coefficient of the filter is adjusted to obtain the gain coefficient required by the average total sound pressure level.

[0066] For example: suppose that information of some noise audio is extracted from historical data, wherein the noise audio information includes the noise audio level and the corresponding total sound pressure level; for example, if noise audios A, B, C, D, E, F, G, H, I, J, K, and L are extracted, the corresponding total sound pressure levels are XA, XB, XC, XD, XE, XF, XG, XH, XI, XJ, XK, and XL respectively; and A, C, H, and J are first-level noises, B, F, I, and K are second-level noises, and D, E, G, and L are third-level noises; in actual operation, more noise audios are extracted for analysis;

[0067] Arrange the noise audio of the same level according to the total sound pressure level, either from small to large or from large to small; for example, the total sound pressure levels of the noise audio of A, C, H, and J are arranged from small to large as XA, XH, XC, and XJ; and divide the total sound pressure level data of the noise audio of A, C, H, and J into two sound pressure groups (XA, XH) and (XC, XJ) with a step size of 2; wherein the step size is selected according to the actual situation and should not be too large to avoid a large gap between the data;

[0068] In this manner, both second- and third-order murmurs are analyzed.

[0069] Calculate the total sound pressure level of the noise audio in the speech to be input; determine whether the total sound pressure level is equal to the average total sound pressure level of the corresponding level; if yes, use the gain coefficient corresponding to the average total sound pressure level as the gain coefficient of the filter of the speech audio to be input; if not, calculate the difference between the total sound pressure level and the average total sound pressure level of the corresponding level, and extract the gain coefficient corresponding to the average total sound pressure level with the smallest difference as the gain coefficient of the filter of the speech audio to be input.

[0070] Step 3: Extract the howling frequency points in the noise-free audio, and eliminate the howling frequency points to obtain the target broadcast voice.

[0071] See also Figure 3Specifically, Fourier transform (such as fast Fourier transform FFT) is applied to convert the audio signal from time domain to frequency domain; in the frequency domain, the spectrum of the audio signal is analyzed to determine the amplitude and phase information of each frequency point; the power spectral density (PSD) describes the power distribution of the signal at different frequencies, and in the power spectral density diagram, the power value corresponding to the required frequency point is found.

[0072] Traverse the frequency points of the noise-free audio from left to right in turn to extract the howling frequency point; mark the frequency point currently being analyzed as the frequency point to be evaluated;

[0073] The power ratio of the frequency point to be evaluated is compared with the power values ​​of other frequency points in turn and the difference is processed. When the power value of another frequency point is smaller than the power value of the frequency point to be evaluated, and the difference between the two is greater than the difference threshold, the frequency point to be evaluated is a howling frequency point, otherwise it is a normal frequency point;

[0074] For example: assuming that the noise of an input voice audio has been eliminated, the power value of each frequency point of the input voice audio is extracted, and the frequency points of the input voice audio are analyzed from left to right, and the first frequency point is analyzed with the second, third, ..., to the last frequency point in turn; assuming that the second frequency point does not meet the condition that the power value is less than the power value of the first frequency point, and the power value difference between the second frequency point and the first frequency point is greater than the difference threshold, and the first and third frequency points meet this condition, then the first frequency point is a howling frequency point; if the third one does not meet the condition either, continue to analyze the first frequency point with other frequency points until the traversal is completed, if the last frequency point is traversed and does not meet this condition, then the first frequency point is not a howling frequency point; at this point, the analysis of the first frequency point is completed, and then the second one is analyzed, and the above analysis method is used until the analysis of the last frequency point is completed, and the howling situation analysis of the entire input voice audio is completed.

[0075] Ratio the amplitude of the howling frequency point to the average amplitude of the normal frequency point, take lg, and then multiply by 20 to get the decibel value of the amplitude of the howling frequency point higher than the normal frequency point;

[0076] The decibel value obtained is summed with the decibel compensation value to obtain the attenuation of the filter, thereby eliminating the howling frequency point.

[0077] It should be noted that the actual attenuation characteristics of the filter may be different from the ideal. For example, the roll-off characteristics of the filter (i.e. the change in attenuation as the frequency changes) may not be completely linear. Therefore, it is necessary to adjust the attenuation of the filter by adding a decibel compensation value to ensure sufficient attenuation at the howling frequency point.

[0078] Among them, a band-stop filter, a low-pass filter or a high-pass filter is selected according to the frequency band width corresponding to the howling frequency point.

[0079] A second aspect of the present invention provides a system for suppressing broadcast howling and eliminating external noise, comprising a noise elimination module, and a noise audio recognition module and a howling suppression module connected thereto;

[0080] The noise audio recognition module is used to identify whether there is noise audio in the voice audio to be input for broadcasting; if yes, the voice audio to be input is the audio with noise, and the noise degree of the audio with noise is classified; if no, the voice audio to be input is the audio without noise;

[0081] The noise elimination module is used to analyze the gain coefficient corresponding to the filter according to the noise level, eliminate the external noise of the broadcast input voice, and obtain noise-free audio;

[0082] The howling suppression module is used to extract the howling frequency points in the noise-free audio and eliminate the howling frequency points to obtain the target broadcast voice.

[0083] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0084] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for suppressing broadcast howling and eliminating external noise, characterized in that: The following steps are involved: Step 1: Identify whether there is noise audio in the input voice audio to be broadcasted; If yes, the audio to be input is noisy audio, and the process goes to step 2; If no, the input voice audio is noise-free audio, and the process goes to step 3; Step 2: classify the noise level of the noisy audio; according to the noise level, analyze the gain coefficient corresponding to the filter, eliminate the external noise of the broadcast input voice, obtain the noise-free audio, and enter step 3; wherein the noise level includes first-level noise, second-level noise and third-level noise; Step 3: Extract the howling frequency points in the noise-free audio, and eliminate the howling frequency points to obtain the target broadcast voice.

2. A method for suppressing broadcast howling and eliminating external noise according to claim 1, characterized in that: The step of identifying whether there is noise audio in the voice audio to be input for broadcasting includes: Extracting input voice audio of a number of broadcasts from historical data; wherein the input voice audio includes normal voice audio and noise audio; The input speech audio is used as the input data of the artificial intelligence model, and the noise audio is used as the output data of the artificial intelligence model; the artificial intelligence model is trained by the input data and the output data to obtain a noise recognition model; The speech audio to be input is input into the noise recognition model, and the noise audio in the speech audio to be input is output.

3. A method for suppressing broadcast howling and eliminating external noise according to claim 2, characterized in that: The classification of the noise level of the noise audio includes: Extract each sound pressure of the noise audio in the input speech audio, marked as Pi; wherein i is the number of the sound pressure, i=0, 1, ..., N, N is a positive integer; The total sound pressure level SYJ of the noise audio is calculated by the formula SYJ=20×lg[(ΣPi) / (N×P0)]; where P0 is the reference sound pressure; The total sound pressure level is compared with the preset range. If it is less than the preset range, it is the first-level noise; if it is within the preset range, it is the second-level noise; if it exceeds the preset range, it is the third-level noise.

4. A method for suppressing broadcast howling and eliminating external noise according to claim 3, characterized in that: The step of analyzing the gain coefficient corresponding to the filter according to the noise level includes: Extracting the total sound pressure level data of several different levels of noise audio from the historical data, and dividing the total sound pressure level of each level of noise audio into several sound pressure groups; wherein the total sound pressure level of each sound pressure group is averaged to obtain the average total sound pressure level of each sound pressure group; Based on the average total sound pressure level, the gain coefficient of the filter is adjusted to obtain the gain coefficient required by the average total sound pressure level.

5. A method for suppressing broadcast howling and eliminating external noise according to claim 4, characterized in that: The total sound pressure level of each level of noise audio is divided into several sound pressure groups, including: The total sound pressure levels of noise audios of the same level are sorted by size, and each of the same step length is grouped together to obtain several sound pressure groups.

6. A method for suppressing broadcast howling and eliminating external noise according to claim 4 or 5, characterized in that: The method of eliminating external noise of the broadcasted voice to be input to obtain noise-free audio includes: Calculate the total sound pressure level of the noise audio in the speech to be input; determine whether the total sound pressure level is equal to the average total sound pressure level of the corresponding level; if yes, use the gain coefficient corresponding to the average total sound pressure level as the gain coefficient of the filter of the speech audio to be input; if not, calculate the difference between the total sound pressure level and the average total sound pressure level of the corresponding level, and extract the gain coefficient corresponding to the average total sound pressure level with the smallest difference as the gain coefficient of the filter of the speech audio to be input.

7. A method for suppressing broadcast howling and eliminating external noise according to claim 6, characterized in that: The step of extracting the howling frequency point in the noise-free audio comprises: Traverse the frequency points of the noise-free audio from left to right in turn to extract the howling frequency point; mark the frequency point currently being analyzed as the frequency point to be evaluated; The power ratio of the frequency point to be evaluated is compared with the power values ​​of other frequency points in turn and the difference is processed. When a power value of another frequency point is smaller than the power value of the frequency point to be evaluated, and the difference between the two is greater than the difference threshold, the frequency point to be evaluated is a howling frequency point.

8. A method for suppressing broadcast howling and eliminating external noise according to claim 7, characterized in that: The step of eliminating the howling frequency point to obtain the target broadcast voice includes: Ratio the amplitude of the howling frequency point to the amplitude of the normal frequency point, take lg, and then multiply by 20 to get the decibel value of the amplitude of the howling frequency point higher than the normal frequency point; The decibel value obtained is summed with the decibel compensation value to obtain the attenuation of the filter, thereby eliminating the howling frequency point.

9. A method for suppressing broadcast howling and eliminating external noise according to claim 8, characterized in that: Select a band-stop filter, a low-pass filter or a high-pass filter according to the frequency band width corresponding to the howling frequency point.

10. A system for suppressing broadcast howling and eliminating external noise, operating based on a method for suppressing broadcast howling and eliminating external noise as claimed in any one of claims 1 to 9, characterized in that: It includes a noise elimination module, and a noise audio recognition module and a howling suppression module connected thereto; The noise audio recognition module is used to recognize whether there is noise audio in the broadcasted voice audio to be input; If yes, the voice audio to be input is a noisy audio, and the noisy audio is classified into noise levels; if no, the voice audio to be input is a noise-free audio; The noise elimination module is used to analyze the gain coefficient corresponding to the filter according to the noise level, eliminate the external noise of the broadcast input voice, and obtain noise-free audio; The howling suppression module is used to extract the howling frequency points in the noise-free audio and eliminate the howling frequency points to obtain the target broadcast voice.

Images