Wireless audio star flash transmission noise management method and system suitable for conference system

Abstract

The invention discloses a wireless audio star flash transmission noise management method and system suitable for a conference system, and the method comprises the steps: obtaining to-be-transmitted audio information, coding the to-be-transmitted audio, carrying out the preliminary separation of the to-be-transmitted audio based on a non-negative matrix factorization algorithm, and obtaining the separated audio information; performing audio classification according to the separated audio information, and classifying to-be-transmitted audios into different audio types to obtain audio classification information; acquiring environment sound audio information according to the audio classification information, forming an environment sound audio sequence, and performing noise reduction management; human voice audio analysis is carried out, the human voice audio is divided into background human voice audio and target human voice audio, noise reduction is carried out on the background human voice audio, and anomaly detection is carried out on the target human voice audio; and carrying out comfort evaluation on the target human voice audio, judging whether human noise exists or not, carrying out audio prediction, and generating noise reduction regulation and control parameters. And the audio transmission noise management quality of the conference scene and the comfort of participants are improved.

Description

Technical Field

[0001] The present invention relates to the field of audio technology, and in particular to a wireless audio star flash transmission noise management method and system suitable for a conference system. Background Art

[0002] In modern conference systems, the widespread application of wireless audio transmission technology has provided great convenience. However, wireless audio transmission will inevitably encounter various noise interference problems, which will not only affect the transmission quality of the audio, but also have a negative impact on the overall experience of the conference. Common sources of wireless audio transmission noise mainly include environmental noise and man-made noise. Environmental noise usually refers to background noise from the surrounding environment, such as wind noise, air conditioning noise, etc.; while man-made noise includes noise caused by human operation, equipment failure, etc.

[0003] In a conference scene, the speaker's voice is uncontrollable during the speech, and the speech sound may suddenly increase. If the sound amplifier fails to control it in time, the amplified speech sound will be played. In a quiet meeting scene, it will bring a bad experience to other people and affect the meeting effect, thus generating artificial noise. Therefore, how to better manage the artificial noise reduction to improve the meeting effect and the comfort of the participants is an important issue. Summary of the invention

[0004] The present invention overcomes the defects of the prior art and provides a wireless audio star flash transmission noise management method and system suitable for a conference system, an important purpose of which is to improve the noise management capability in a conference scene and the comfort of participants.

[0005] To achieve the above-mentioned purpose, the first aspect of the present invention provides a wireless audio star flash transmission noise management method applicable to a conference system, comprising: Obtain the audio information to be transmitted of the target conference scene, encode the audio to be transmitted, and perform preliminary separation of the audio to be transmitted based on the non-negative matrix decomposition algorithm to obtain separated audio information; Constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into ambient sound audio and human voice audio, and obtaining audio classification information; Acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attributes of the ambient sound audio, and performing noise reduction management on the ambient sound audio; Acquire human voice audio information according to the audio classification information, perform human voice audio analysis, divide the human voice audio into background human voice audio and target human voice audio, perform noise reduction on the background human voice audio, and perform anomaly detection on the target human voice audio; Based on the anomaly detection results, the comfort level of the target vocal audio is evaluated, whether there is artificial noise is determined and audio prediction is performed, and noise reduction control parameters are generated to manage the noise reduction of the target vocal audio.

[0006] In this solution, the audio information to be transmitted of the target conference scene is obtained, the audio to be transmitted is encoded, and the audio to be transmitted is preliminarily separated based on the non-negative matrix decomposition algorithm to obtain the separated audio information, which specifically includes: The audio of the target conference scene is collected through an audio recording device to obtain the audio information to be transmitted of the target conference scene; Preset a time sequence, encode the audio to be transmitted according to the audio information to be transmitted, embed the encoded audio data into the time sequence, perform data preprocessing, and obtain mixed audio information; Introducing a short-time Fourier transform algorithm to perform frame processing on the mixed audio information, windowing the audio data of each frame based on a Hanning window function, and multiplying each frame of audio data by the Hanning window function to obtain windowed audio information; Performing discrete Fourier transform on the mixed audio data of each time window according to the windowed audio information, obtaining the frequency spectrum corresponding to the mixed audio data of each time window, and forming a time-frequency matrix; Decomposing the time-frequency matrix based on a non-negative matrix decomposition algorithm, randomly initializing a basis matrix and an activation coefficient matrix, taking KL divergence as an optimization target, presetting a loss function and adopting a multiplication update rule to iteratively optimize the basis matrix and the activation coefficient matrix; Obtain the optimized basis matrix and activation coefficient matrix, extract the basis vector and activation coefficient vector respectively, obtain the spectral feature representation corresponding to the ambient sound and human voice through data retrieval, and classify the basis vector and activation coefficient vector into the ambient sound vector category and human voice vector category respectively by combining the clustering algorithm; Based on the classification results, the time-frequency representation matrix of the ambient sound and human voice is calculated, and an inverse short-time Fourier transform is performed to obtain the separated audio information.

[0007] In this solution, the audio classification model is constructed, audio classification is performed according to the separated audio information, the audio to be transmitted is classified into different audio types, and audio classification information is obtained, which specifically includes: Acquire separated audio information, use a Mel filter to perform feature extraction on the separated audio information, obtain a Mel frequency based on the Mel filter, and perform filtering processing on the separated audio information; Perform logarithmic compression on the separated audio information after filtering to obtain the separated audio information after the logarithmic compression result, perform discrete cosine transform and generate audio cepstrum coefficients; The generated audio cepstral coefficients are compared with a preset threshold, and the audio cepstral coefficients greater than the preset threshold are selected and merged to obtain separated audio feature information; Preset classification categories, obtain spectrograms corresponding to each category through big data retrieval, and divide them according to the spectral entropy of the spectrograms of each category, obtain the voiceprint features corresponding to the spectrograms of each category according to the division results and associate them to form a first data set; Constructing an audio classification model, performing model training through the first data set to obtain an audio classification model that meets expectations, and inputting the separated audio feature information into the audio classification model to perform audio classification; Calculating the spectral entropy of the separated audio based on the input separated audio feature information, obtaining the spectrogram and voiceprint features of the corresponding category according to the spectral entropy of the separated audio, generating the separated audio spectrogram and the separated audio voiceprint features through the separated audio information, and performing similarity calculation to obtain a similarity value; The calculated similarity value is compared with a preset threshold, and the classification result of the separated audio is output according to the judgment result to obtain audio classification information.

[0008] In this solution, the step of acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and performing noise reduction management on the ambient sound audio specifically includes: Acquire audio classification information, extract ambient sound audio information according to the audio classification information, form an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and extract human voice audio information through the audio classification information to form a human voice audio sequence; Based on the ambient sound audio sequence, an average spectrum of the noise sequence is calculated as the noise spectrum, the ambient sound audio sequence is time-aligned with the human voice audio sequence, and a sequence where the human voice and the ambient sound overlap is marked to obtain a marked sequence; Extracting features of the ambient sound audio information to obtain frequency, amplitude and audio waveform features, and performing periodic analysis to obtain periodic analysis information; The spectrum of the noisy frequency is calculated according to the marker sequence, and subtracted from the noise spectrum to obtain the subtracted spectrum. The correction parameters are generated through the periodic analysis information, and the subtracted spectrum is corrected to obtain the corrected spectrum. The inverse Fourier transform is used to generate the ambient sound audio noise reduction result information.

[0009] In this solution, the human voice audio information is obtained according to the audio classification information, the human voice audio is analyzed, the human voice audio is divided into background human voice audio and target human voice audio, the background human voice audio is denoised, and the target human voice audio is detected for abnormalities, specifically including: Acquire mixed human voice audio based on historical data retrieval, acquire a spectrogram of the mixed human voice audio according to the acquired mixed human voice audio, and divide the mixed human voice audio according to frequency to obtain first division information; Obtain the frequency and sound features of the background human voice and the target human voice in the speaking scene through instance statistical analysis, and divide the mixed human voice audio into the background human voice and the target human voice in combination with the first division information to form an instance data set; Acquire audio classification information, acquire human voice audio information according to the audio classification information, perform pre-emphasis filtering on the human voice audio information, and perform feature extraction to acquire human voice audio feature information; Constructing a human voice audio analysis model based on SVM, performing model training through the example data set to obtain a human voice audio analysis model that meets expectations, inputting the human voice audio feature information into the human voice audio analysis model for analysis, and obtaining human voice audio analysis information; The human voice audio analysis information divides the human voice audio into background human voice and target human voice, and uses spectrum subtraction to reduce background human voice noise; Extract target human voice audio based on human voice audio analysis information, build an anomaly detection model through generative adversarial network, input target human voice audio and generate original human voice audio sequence according to target human voice audio features, and obtain generated human voice audio sequence through generator; A reconstruction error between the original human voice audio sequence and the generated human voice audio sequence is calculated, and abnormal human voice audio detection is performed according to the reconstruction error to obtain abnormal human voice audio detection information.

[0010] In this solution, the comfort level of the target human voice audio is evaluated based on the abnormal detection result, whether there is artificial noise and audio prediction is performed, and noise reduction control parameters are generated to perform noise reduction management on the target human voice audio, specifically including: Based on data retrieval, the audio of the speaker's voice under each amplifier standard in different conference scenarios is obtained, and the comfort level of each speaker's voice audio is evaluated through expert analysis to obtain the audio of the speaker's voice with different comfort levels; Extracting human voice audio features corresponding to human voice audio with different comfort levels to construct comfort evaluation rules, obtaining abnormal human voice audio detection information, and performing comfort evaluation on abnormal human voice audio to obtain comfort evaluation information; If the comfort evaluation result is that the target human voice audio at the current moment is comfortable, then voice enhancement is performed on the target human voice audio at the current moment; If the comfort evaluation result is that the target human voice audio at the current moment is uncomfortable, then artificial noise appears in the target human voice audio at the current moment, noise management is performed, and abnormal human voice audio features are extracted according to the abnormal human voice audio detection information; Obtaining a comfortable audio benchmark according to the comfort evaluation rule, performing deviation calculation on the extracted abnormal human voice audio feature to obtain a deviation value, performing noise reduction control on the abnormal human voice audio based on the calculated deviation value, and generating a noise reduction control parameter; The Markov algorithm is introduced to predict the audio according to the extracted abnormal human voice audio features, the initial state is set according to the abnormal human voice audio features, and the state transfer matrix of the initial state is retrieved in the preset state space; Analyze the state transition probability of the next state of the abnormal human voice audio at the current moment according to the state transition matrix, and generate a state transition dynamic sequence of the abnormal human voice audio at the current moment; Acquire human voice audio analysis information, extract target human voice audio features and generate the actual state change sequence of the target human voice, analyze the state transition probability of the actual audio change, select the state transition sequence that conforms to the actual audio change in combination with the state transition dynamic sequence, and obtain the final state transition sequence; Generating an audio prediction result according to the final state transition sequence, performing anomaly detection and comfort evaluation on the audio prediction result, and performing noise reduction control on the audio prediction result whose evaluation result is uncomfortable, to generate a noise reduction control prediction parameter; The target human voice audio at the next moment is obtained for feature extraction and the audio state change sequence at the next moment is generated, which is compared and analyzed with the final state transition sequence, and the noise reduction control prediction parameters are corrected according to the comparison and analysis results.

[0011] A second aspect of the present invention provides a wireless audio star flash transmission noise management system applicable to a conference system, the system comprising: a memory, a processor, the memory containing a wireless audio star flash transmission noise management method program applicable to the conference system, the wireless audio star flash transmission noise management method program applicable to the conference system when executed by the processor implements the following steps: Obtain the audio information to be transmitted of the target conference scene, encode the audio to be transmitted, and perform preliminary separation of the audio to be transmitted based on the non-negative matrix decomposition algorithm to obtain separated audio information; Constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; Acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attributes of the ambient sound audio, and performing noise reduction management on the ambient sound audio; Acquire human voice audio information according to the audio classification information, perform human voice audio analysis, divide the human voice audio into background human voice audio and target human voice audio, perform noise reduction on the background human voice audio, and perform anomaly detection on the target human voice audio; Based on the anomaly detection results, the comfort level of the target vocal audio is evaluated, whether there is artificial noise is determined and audio prediction is performed, and noise reduction control parameters are generated to manage the noise reduction of the target vocal audio.

[0012] The present invention discloses a wireless audio star flash transmission noise management method and system suitable for conference systems, including: obtaining audio information to be transmitted, encoding the audio to be transmitted, performing preliminary separation on the audio to be transmitted based on a non-negative matrix decomposition algorithm to obtain separated audio information; performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; obtaining ambient sound audio information according to the audio classification information and forming an ambient sound audio sequence, and performing noise reduction management; performing human voice audio analysis, dividing the human voice audio into background human voice audio and target human voice audio, performing noise reduction on the background human voice audio, and performing abnormality detection on the target human voice audio; performing comfort evaluation on the target human voice audio, judging whether there is artificial noise and performing audio prediction, and generating noise reduction control parameters. The audio transmission noise management quality of the conference scene and the comfort of the participants are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions 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 according to the drawings without paying creative work.

[0014] Figure 1 A flow chart of a wireless audio star flash transmission noise management method applicable to a conference system provided by an embodiment of the present invention; Figure 2 A flowchart of noise reduction management for target human voice audio provided by an embodiment of the present invention; Figure 3 A block diagram of a wireless audio star flash transmission noise management system applicable to a conference system provided by an embodiment of the present invention; The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0016] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flow chart of a wireless audio star flash transmission noise management method applicable to a conference system provided by an embodiment of the present invention; like Figure 1 As shown, the present invention provides a flow chart of a wireless audio star flash transmission noise management method applicable to a conference system, including: S102, obtaining audio information to be transmitted of the target conference scene, encoding the audio to be transmitted, and performing preliminary separation of the audio to be transmitted based on a non-negative matrix decomposition algorithm to obtain separated audio information; S104, constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; S106, acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and performing noise reduction management on the ambient sound audio; S108, acquiring human voice audio information according to the audio classification information, performing human voice audio analysis, dividing the human voice audio into background human voice audio and target human voice audio, performing noise reduction on the background human voice audio, and performing abnormality detection on the target human voice audio; S110, based on the abnormality detection result, the comfort level of the target human voice audio is evaluated, whether there is artificial noise and audio prediction is performed, and noise reduction control parameters are generated to perform noise reduction management on the target human voice audio.

[0018] It should be noted that the present invention provides a wireless audio star flash transmission noise management method and system suitable for conference systems, which encodes the audio information to be transmitted, uses a non-negative matrix decomposition algorithm to perform preliminary separation of the audio to be transmitted, and performs audio classification according to the classification results, and further subdivides the audio to be transmitted into different audio types. After audio classification, an ambient sound audio sequence is formed based on the temporal attributes of the ambient sound audio, and noise reduction management is performed on the ambient sound audio. According to the audio classification information, human voice audio information is obtained, human voice audio analysis is performed, the human voice audio is divided into target human voice and background human voice, the background human voice audio is noise-reduced, and the target human voice audio is detected for abnormality to analyze the abnormal audio features present in the target human voice audio. Finally, the comfort level of the target human voice audio is evaluated based on the abnormality detection results, and it is determined whether there is artificial noise and audio prediction is performed. The noise control prediction parameters are generated by the audio prediction results to improve the audio noise reduction efficiency, thereby improving the audio transmission effect and rate. While improving the audio transmission quality of the conference scene, avoid affecting the conference effect due to artificial noise.

[0019] Further, in a preferred embodiment of the present invention, the step of obtaining the audio information to be transmitted of the target conference scene, encoding the audio to be transmitted, and performing preliminary separation of the audio to be transmitted based on a non-negative matrix decomposition algorithm to obtain separated audio information specifically includes: The audio of the target conference scene is collected through an audio recording device to obtain the audio information to be transmitted of the target conference scene; Preset a time sequence, encode the audio to be transmitted according to the audio information to be transmitted, embed the encoded audio data into the time sequence, perform data preprocessing, and obtain mixed audio information; Introducing a short-time Fourier transform algorithm to perform frame processing on the mixed audio information, windowing the audio data of each frame based on a Hanning window function, and multiplying each frame of audio data by the Hanning window function to obtain windowed audio information; Performing discrete Fourier transform on the mixed audio data of each time window according to the windowed audio information, obtaining the frequency spectrum corresponding to the mixed audio data of each time window, and forming a time-frequency matrix; Decomposing the time-frequency matrix based on a non-negative matrix decomposition algorithm, randomly initializing a basis matrix and an activation coefficient matrix, taking KL divergence as an optimization target, presetting a loss function and adopting a multiplication update rule to iteratively optimize the basis matrix and the activation coefficient matrix; Obtain the optimized basis matrix and activation coefficient matrix, extract the basis vector and activation coefficient vector respectively, obtain the spectral feature representation corresponding to the ambient sound and human voice through data retrieval, and classify the basis vector and activation coefficient vector into the ambient sound vector category and human voice vector category respectively by combining the clustering algorithm; Based on the classification results, the time-frequency representation matrix of the ambient sound and human voice is calculated, and an inverse short-time Fourier transform is performed to obtain the separated audio information.

[0020] It should be noted that the audio recording device is used to collect audio of the target conference scene, obtain the audio information to be transmitted, encode it, and embed it into the preset time series to form mixed audio information. Next, the short-time Fourier transform (STFT) algorithm is introduced to perform frame processing on the mixed audio information. The audio data of each frame is windowed based on the Hanning window function, that is, each frame of audio data is multiplied by the Hanning window function to obtain the windowed audio information. According to the windowed audio information, the mixed audio data of each time window is subjected to discrete Fourier transform (DFT) to obtain the spectrum representation of the mixed audio data of each time window and form a time-frequency matrix. At this time, the time-frequency matrix is ​​decomposed based on the non-negative matrix decomposition (NMF) algorithm. The basis matrix and the activation coefficient matrix are randomly initialized, and then the KL divergence is used as the optimization target, the loss function is preset, and the multiplication update rule is used to iteratively optimize the basis matrix and the activation coefficient matrix to obtain the optimized basis matrix and the activation coefficient matrix. The basis vector and the activation coefficient vector are extracted respectively, and the spectrum feature representation corresponding to the ambient sound and the human voice is obtained through data retrieval. Combined with the clustering algorithm, the basis vector and activation coefficient vector are classified into environmental sound vector category and human voice vector category. Finally, the time-frequency representation matrix of environmental sound and human voice is calculated based on the classification results, and the inverse short-time Fourier transform (ISTFT) is performed to obtain the separated audio information, realizing the effective separation of environmental sound and human voice in the mixed audio signal, providing a basis for subsequent analysis.

[0021] Further, in a preferred embodiment of the present invention, the step of constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information specifically includes: Acquire separated audio information, use a Mel filter to perform feature extraction on the separated audio information, obtain a Mel frequency based on the Mel filter, and perform filtering processing on the separated audio information; Perform logarithmic compression on the separated audio information after filtering to obtain the separated audio information after the logarithmic compression result, perform discrete cosine transform and generate audio cepstrum coefficients; The generated audio cepstral coefficients are compared with a preset threshold, and the audio cepstral coefficients greater than the preset threshold are selected and merged to obtain separated audio feature information; Preset classification categories, obtain spectrograms corresponding to each category through big data retrieval, and divide them according to the spectral entropy of the spectrograms of each category, obtain the voiceprint features corresponding to the spectrograms of each category according to the division results and associate them to form a first data set; Constructing an audio classification model, performing model training through the first data set to obtain an audio classification model that meets expectations, and inputting the separated audio feature information into the audio classification model to perform audio classification; Calculating the spectral entropy of the separated audio based on the input separated audio feature information, obtaining the spectrogram and voiceprint features of the corresponding category according to the spectral entropy of the separated audio, generating the separated audio spectrogram and the separated audio voiceprint features through the separated audio information, and performing similarity calculation to obtain a similarity value; The calculated similarity value is compared with a preset threshold, and the classification result of the separated audio is output according to the judgment result to obtain audio classification information.

[0022] It should be noted that after the audio to be transmitted is initially separated, a mixed audio of ambient sound and human voice is obtained. At this time, further classification is required to understand what type of sound the corresponding audio is, such as background human voice, target human voice, equipment noise, etc. A Mel filter bank is used to extract features from the separated audio. The Mel filter bank simulates the nonlinear perception of frequency by the human ear, converts the audio signal to the Mel frequency scale, and performs filtering processing to highlight the frequency range that is more sensitive to human hearing. The audio information after filtering is logarithmically compressed. Logarithmic compression can reduce the dynamic range in the audio signal and make subsequent processing more stable. The logarithmically compressed audio signal is converted from the time domain to the frequency domain through discrete cosine transform (DCT), and the audio cepstral coefficients (MFCCs) are generated. MFCCs can well represent the spectral envelope of the audio. Next, the MFCCs are compared with the preset threshold, and the coefficients greater than the threshold are selected for merging to remove unnecessary noise or interference to obtain the separated audio feature information. At the same time, the spectrograms of various categories are retrieved through big data, and the spectrograms are divided based on the spectral entropy to construct the first data set. Spectral entropy is an indicator to measure the complexity or uncertainty of a signal, and is used here to distinguish different categories of audio features. Based on the division results, the voiceprint features corresponding to the spectrograms of each category are extracted and associated. The voiceprint features represent the unique identification marks of different audio categories. The audio classification model is trained using the first data set, and the category to which the new audio signal belongs is identified by learning the mapping relationship between the voiceprint features and the audio categories. The separated audio feature information is input into the model to obtain the classification results of the separated audio. This provides a basis for subsequent noise management.

[0023] Further, in a preferred embodiment of the present invention, the step of acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the temporal properties of the ambient sound audio, and performing noise reduction management on the ambient sound audio specifically includes: Acquire audio classification information, extract ambient sound audio information according to the audio classification information, form an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and extract human voice audio information through the audio classification information to form a human voice audio sequence; Based on the ambient sound audio sequence, an average spectrum of the noise sequence is calculated as the noise spectrum, the ambient sound audio sequence is time-aligned with the human voice audio sequence, and a sequence where the human voice and the ambient sound overlap is marked to obtain a marked sequence; Extracting features of the ambient sound audio information to obtain frequency, amplitude and audio waveform features, and performing periodic analysis to obtain periodic analysis information; The spectrum of the noisy frequency is calculated according to the marker sequence, and subtracted from the noise spectrum to obtain the subtracted spectrum. The correction parameters are generated through the periodic analysis information, and the subtracted spectrum is corrected to obtain the corrected spectrum. The inverse Fourier transform is used to generate the ambient sound audio noise reduction result information.

[0024] It should be noted that in a conference scenario, in order to ensure the audio transmission quality of the speaker of the conference content, the ambient sound audio collected by the audio collector is regarded as ambient noise. When separating the ambient noise from the audio to be transmitted, it is necessary to perform noise reduction management on it. The ambient noise can be reduced to a certain extent through spectral subtraction, thereby avoiding the listening effect of the participants affected by the ambient noise, so as to improve the meeting quality of the target conference scenario.

[0025] Further, in a preferred embodiment of the present invention, the human voice audio information is obtained according to the audio classification information, the human voice audio is analyzed, the human voice audio is divided into background human voice audio and target human voice audio, the background human voice audio is denoised, and the target human voice audio is detected for abnormalities, specifically including: Acquire mixed human voice audio based on historical data retrieval, acquire a spectrogram of the mixed human voice audio according to the acquired mixed human voice audio, and divide the mixed human voice audio according to frequency to obtain first division information; Obtain the frequency and sound features of the background human voice and the target human voice in the speaking scene through instance statistical analysis, and divide the mixed human voice audio into the background human voice and the target human voice in combination with the first division information to form an instance data set; Acquire audio classification information, acquire human voice audio information according to the audio classification information, perform pre-emphasis filtering on the human voice audio information, and perform feature extraction to acquire human voice audio feature information; Constructing a human voice audio analysis model based on SVM, performing model training through the example data set to obtain a human voice audio analysis model that meets expectations, inputting the human voice audio feature information into the human voice audio analysis model for analysis, and obtaining human voice audio analysis information; The human voice audio analysis information divides the human voice audio into background human voice and target human voice, and uses spectrum subtraction to reduce background human voice noise; Extract target human voice audio based on human voice audio analysis information, build an anomaly detection model through generative adversarial network, input target human voice audio and generate original human voice audio sequence according to target human voice audio features, and obtain generated human voice audio sequence through generator; A reconstruction error between the original human voice audio sequence and the generated human voice audio sequence is calculated, and abnormal human voice audio detection is performed according to the reconstruction error to obtain abnormal human voice audio detection information.

[0026] It should be noted that, first, based on the retrieval of historical data, an audio file containing a mixture of multiple human voices is obtained and a spectrogram is generated to show the frequency components of the audio signal changing over time. The mixed human voice audio is preliminarily divided based on the natural separation characteristics of different human voice components in the audio signal in the frequency domain through the spectrogram to obtain the first division information. Next, through example statistical analysis, the frequency distribution and sound characteristics of the background human voice and the target human voice in the speaking scene are understood. Combined with the first division information, the mixed human voice audio is divided into two parts: background human voice and target human voice, and an example data set is formed. A human voice audio analysis model is constructed using a support vector machine (SVM), and the human voice audio feature information is input into the human voice audio analysis model for analysis, and the audio is divided into background human voice and target human voice. At this time, the background human voice is human voice noise, which will greatly affect the audio quality of the speaker, and the spectrum subtraction technology is used to reduce the impact of the background human voice on the target human voice. Finally, the generative adversarial network (GAN) is used to build an anomaly detection model, and the target human voice audio is extracted based on the human voice audio analysis information for anomaly detection, and the abnormal audio features in the target human voice are detected, including frequency, amplitude, etc. The abnormal human voice audio detection information is output to provide valuable reference for subsequent audio processing or analysis.

[0027] Further, in a preferred embodiment of the present invention, the comfort level of the target human voice audio is evaluated based on the abnormality detection result, whether there is artificial noise and audio prediction is performed, and noise reduction control parameters are generated to perform noise reduction management on the target human voice audio, specifically including: Based on data retrieval, the audio of the speaker's voice under each amplifier standard in different conference scenarios is obtained, and the comfort level of each speaker's voice audio is evaluated through expert analysis to obtain the audio of the speaker's voice with different comfort levels; Extracting human voice audio features corresponding to human voice audio with different comfort levels to construct comfort evaluation rules, obtaining abnormal human voice audio detection information, and performing comfort evaluation on abnormal human voice audio to obtain comfort evaluation information; If the comfort evaluation result is that the target human voice audio at the current moment is comfortable, then voice enhancement is performed on the target human voice audio at the current moment; If the comfort evaluation result is that the target human voice audio at the current moment is uncomfortable, then artificial noise appears in the target human voice audio at the current moment, noise management is performed, and abnormal human voice audio features are extracted according to the abnormal human voice audio detection information; Obtaining a comfortable audio benchmark according to the comfort evaluation rule, performing deviation calculation on the extracted abnormal human voice audio feature to obtain a deviation value, performing noise reduction control on the abnormal human voice audio based on the calculated deviation value, and generating a noise reduction control parameter; The Markov algorithm is introduced to predict the audio according to the extracted abnormal human voice audio features, the initial state is set according to the abnormal human voice audio features, and the state transfer matrix of the initial state is retrieved in the preset state space; Analyze the state transition probability of the next state of the abnormal human voice audio at the current moment according to the state transition matrix, and generate a state transition dynamic sequence of the abnormal human voice audio at the current moment; Acquire human voice audio analysis information, extract target human voice audio features and generate the actual state change sequence of the target human voice, analyze the state transition probability of the actual audio change, select the state transition sequence that conforms to the actual audio change in combination with the state transition dynamic sequence, and obtain the final state transition sequence; Generating an audio prediction result according to the final state transition sequence, performing anomaly detection and comfort evaluation on the audio prediction result, and performing noise reduction control on the audio prediction result whose evaluation result is uncomfortable, to generate a noise reduction control prediction parameter; The target human voice audio at the next moment is obtained for feature extraction and the audio state change sequence at the next moment is generated, which is compared and analyzed with the final state transition sequence, and the noise reduction control prediction parameters are corrected according to the comparison and analysis results.

[0028] It should be noted that in the conference scene, due to the uncontrollability of the speaker's speech sound, the speaker deliberately or unintentionally increases the volume of the speech. At this time, if the audio is transmitted and output completely according to the audio at the time of collection, it may cause discomfort to the participants, thereby affecting the quality of the meeting. Through data retrieval, the speaking voice audio samples from different conference scenes that follow the standards of various amplifiers are collected. Subsequently, the expert analysis method is used to evaluate the comfort of the speaking voice audio samples. The human voice audio features corresponding to the speaking voice audio of different comfort levels are extracted, and the comfort evaluation rules are constructed. It is used for the subsequent comfort evaluation of abnormal human voice audio and also as a benchmark for noise reduction control. When abnormal human voice audio is detected, the comfort evaluation is immediately performed to determine whether noise reduction measures need to be taken. If the comfort evaluation result shows that the target human voice audio at the current moment is comfortable, it is subjected to speech enhancement processing to improve the clarity and intelligibility of the audio. On the contrary, if the evaluation result is uncomfortable, it is determined that there is artificial noise in the current target human voice audio, and noise reduction management is performed. The benchmark of comfortable audio is obtained according to the comfort evaluation rule. By comparing this benchmark with the extracted abnormal audio features, the deviation value between them is calculated, and noise reduction processing is performed based on the noise reduction control parameters corresponding to the deviation value. At the same time, in order to improve the efficiency and effect of audio transmission, after detecting the abnormal human voice audio, the subsequent audio information is predicted, and the Markov algorithm is used for audio prediction. First, the initial state is set according to the characteristics of the abnormal human voice audio, and the state transition matrix of the initial state is retrieved in the preset state space. By analyzing the state transition matrix, the state transition probability of the current abnormal human voice audio at the next moment is predicted, and the corresponding state transition dynamic sequence is generated. At the same time, the human voice audio analysis information is obtained to extract the characteristics of the target human voice audio and generate its actual state change sequence. By comparing the state transition probability of the actual audio change with the predicted state transition dynamic sequence, the state transition sequence that best matches the actual audio change is selected as the final state transition sequence. The audio prediction result is generated based on the final state transition sequence. Subsequently, abnormal detection and comfort evaluation are performed, and noise reduction control is performed on the prediction result with the evaluation result of discomfort, and the corresponding noise reduction control prediction parameters are generated. Finally, when the target human voice audio at the next moment is obtained, feature extraction is performed on it and a new audio state change sequence is generated. Through comparative analysis of the final state transition sequence, the noise reduction control prediction parameters are corrected to ensure the continued effectiveness and accuracy of the noise reduction process, thereby avoiding over-control.

[0029] Figure 2 A flowchart of noise reduction management for target human voice audio provided by an embodiment of the present invention; like Figure 2 As shown, the present invention provides a flow chart for noise reduction management of target human voice audio, including: S202, extracting target human voice audio based on human voice audio analysis information, and inputting the target human voice audio into an abnormality detection model to perform abnormal human voice audio detection, thereby obtaining abnormal human voice audio detection information; S204, based on the comfort evaluation rule and in combination with the abnormal human voice audio detection information, performing comfort evaluation on the abnormal human voice audio to obtain comfort evaluation information; S206, if the comfort evaluation result is that the target human voice audio at the current moment is comfortable, voice enhancement is performed on the target human voice audio at the current moment; S208, if the comfort evaluation result is that the target human voice audio at the current moment is uncomfortable, noise reduction management is performed on the target human voice audio at the current moment, and abnormal human voice audio features are extracted according to the abnormal human voice audio detection information; S210, obtaining a comfortable audio reference according to a comfort evaluation rule, performing deviation calculation on the extracted abnormal human voice audio feature to obtain a deviation value, and performing noise reduction control on the abnormal human voice audio based on the calculated deviation value to generate a noise reduction control parameter; S212, performing audio prediction based on the extracted abnormal human voice audio features, performing abnormality detection and comfort evaluation on the audio prediction results, and performing noise reduction control on the audio prediction results whose evaluation results are uncomfortable, to generate noise reduction control prediction parameters; S214, obtaining the target human voice audio at the next moment for feature extraction and generating the audio state change sequence at the next moment, performing comparative analysis on the state transition sequence corresponding to the noise reduction control prediction parameter and correcting the noise reduction control parameter.

[0030] In addition, in a wireless audio star flash transmission noise management method applicable to a conference system provided by the present invention, the following steps are also included: Based on historical data retrieval, target human voice audio samples with artificial noise in the conference scene and corresponding audio noise reduction control parameters are obtained, audio sample features corresponding to the target human voice audio samples are extracted, and audio sample feature information is obtained; Calculating the Euclidean distance between the audio samples according to the audio sample feature information, classifying the audio samples by using the k-means clustering algorithm, and associating the corresponding audio noise reduction control parameters to obtain audio sample classification information; Combining the audio sample classification information and the audio sample feature information, constructing an audio feature portrait of each category of audio samples, and forming an audio noise reduction management database based on the constructed audio sample feature portrait and the corresponding audio noise reduction control parameters; Acquire real-time target human voice audio information, perform similarity calculation between the real-time target human voice audio information and the audio feature portrait of the audio noise reduction management database, and obtain a similarity value; The similarity value is judged with a preset threshold, a corresponding audio sample category is selected according to the judgment result, and the similarity between the real-time target human voice audio information and each audio sample in the target audio sample category is calculated, and each audio sample in the target audio sample category is sorted according to the calculated similarity; Selecting a final audio sample based on the sorting result, and using the audio noise reduction control parameters corresponding to the final audio sample as control parameters of the real-time target human voice audio to perform noise reduction management; If, after the similarity calculation, there is no audio feature portrait similar to the real-time target human voice audio information in the audio noise reduction management database, an audio noise reduction control parameter analysis is performed on the real-time target human voice audio, and a new category of audio samples is generated based on the audio features of the real-time target human voice audio, and the audio noise reduction management database is updated in combination with the corresponding audio noise reduction control parameters.

[0031] It should be noted that in order to improve the efficiency of audio transmission noise reduction management, the target human voice audio samples with artificial noise in the conference scene are analyzed through historical data, and then the audio feature portraits of each category of audio samples are constructed, and an audio noise reduction management database is formed based on the constructed audio sample feature portraits and the corresponding audio noise reduction control parameters. The real-time target human voice audio is noise-reduced through the audio noise reduction management database to improve the analysis and noise reduction efficiency, thereby improving the quality of audio transmission. In addition, when there is no audio feature portrait similar to the real-time target human voice audio information in the audio noise reduction management database, a new audio sample category is generated according to the real-time target human voice audio, and the corresponding audio noise reduction control parameters are associated to update the audio noise reduction management database to improve the adaptability and extensiveness of the audio noise reduction management database.

[0032] Figure 3 A wireless audio star flash transmission noise management system 3 applicable to a conference system is provided in one embodiment of the present invention. The system includes: a memory 31 and a processor 32. The memory 31 contains a wireless audio star flash transmission noise management method program applicable to the conference system. When the wireless audio star flash transmission noise management method program applicable to the conference system is executed by the processor 32, the following steps are implemented: Obtain the audio information to be transmitted of the target conference scene, encode the audio to be transmitted, and perform preliminary separation of the audio to be transmitted based on the non-negative matrix decomposition algorithm to obtain separated audio information; Constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; Acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attributes of the ambient sound audio, and performing noise reduction management on the ambient sound audio; Acquire human voice audio information according to the audio classification information, perform human voice audio analysis, divide the human voice audio into background human voice audio and target human voice audio, perform noise reduction on the background human voice audio, and perform anomaly detection on the target human voice audio; Based on the anomaly detection results, the comfort level of the target vocal audio is evaluated, whether there is artificial noise is determined and audio prediction is performed, and noise reduction control parameters are generated to manage the noise reduction of the target vocal audio.

[0033] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0034] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0035] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0036] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0037] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0038] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A wireless audio star flash transmission noise management method suitable for a conference system, characterized in that: include: Obtain the audio information to be transmitted of the target conference scene, encode the audio to be transmitted, and perform preliminary separation of the audio to be transmitted based on the non-negative matrix decomposition algorithm to obtain separated audio information; Constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; Acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attributes of the ambient sound audio, and performing noise reduction management on the ambient sound audio; Acquire human voice audio information according to the audio classification information, perform human voice audio analysis, divide the human voice audio into background human voice audio and target human voice audio, perform noise reduction on the background human voice audio, and perform anomaly detection on the target human voice audio; Based on the anomaly detection results, the comfort level of the target vocal audio is evaluated, whether there is artificial noise is determined and audio prediction is performed, and noise reduction control parameters are generated to manage the noise reduction of the target vocal audio.

2. According to claim 1, a wireless audio star flash transmission noise management method suitable for a conference system is characterized in that: The step of obtaining the audio information to be transmitted of the target conference scene, encoding the audio to be transmitted, and performing preliminary separation of the audio to be transmitted based on a non-negative matrix decomposition algorithm to obtain separated audio information specifically includes: The audio of the target conference scene is collected through an audio recording device to obtain the audio information to be transmitted of the target conference scene; Preset a time sequence, encode the audio to be transmitted according to the audio information to be transmitted, embed the encoded audio data into the time sequence, perform data preprocessing, and obtain mixed audio information; Introducing a short-time Fourier transform algorithm to perform frame processing on the mixed audio information, windowing the audio data of each frame based on a Hanning window function, and multiplying each frame of audio data by the Hanning window function to obtain windowed audio information; Performing discrete Fourier transform on the mixed audio data of each time window according to the windowed audio information, obtaining the frequency spectrum corresponding to the mixed audio data of each time window, and forming a time-frequency matrix; Decomposing the time-frequency matrix based on a non-negative matrix decomposition algorithm, randomly initializing a basis matrix and an activation coefficient matrix, taking KL divergence as an optimization target, presetting a loss function and adopting a multiplication update rule to iteratively optimize the basis matrix and the activation coefficient matrix; Obtain the optimized basis matrix and activation coefficient matrix, extract the basis vector and activation coefficient vector respectively, obtain the spectral feature representation corresponding to the ambient sound and human voice through data retrieval, and classify the basis vector and activation coefficient vector into the ambient sound vector category and human voice vector category respectively by combining the clustering algorithm; Based on the classification results, the time-frequency representation matrix of the ambient sound and human voice is calculated, and an inverse short-time Fourier transform is performed to obtain the separated audio information.

3. According to claim 1, a wireless audio star flash transmission noise management method suitable for a conference system is characterized in that: The audio classification model is constructed, audio classification is performed according to the separated audio information, and the audio to be transmitted is classified into different audio types to obtain audio classification information, specifically including: Acquire separated audio information, use a Mel filter to perform feature extraction on the separated audio information, obtain a Mel frequency based on the Mel filter, and perform filtering processing on the separated audio information; Perform logarithmic compression on the separated audio information after filtering to obtain the separated audio information after the logarithmic compression result, perform discrete cosine transform and generate audio cepstrum coefficients; The generated audio cepstral coefficients are compared with a preset threshold, and the audio cepstral coefficients greater than the preset threshold are selected and merged to obtain separated audio feature information; Preset classification categories, obtain spectrograms corresponding to each category through big data retrieval, and divide them according to the spectral entropy of the spectrograms of each category, obtain the voiceprint features corresponding to the spectrograms of each category according to the division results and associate them to form a first data set; Constructing an audio classification model, performing model training through the first data set to obtain an audio classification model that meets expectations, and inputting the separated audio feature information into the audio classification model to perform audio classification; Calculating the spectral entropy of the separated audio based on the input separated audio feature information, obtaining the spectrogram and voiceprint features of the corresponding category according to the spectral entropy of the separated audio, generating the separated audio spectrogram and the separated audio voiceprint features through the separated audio information, and performing similarity calculation to obtain a similarity value; The calculated similarity value is compared with a preset threshold, and the classification result of the separated audio is output according to the judgment result to obtain audio classification information.

4. According to claim 1, a wireless audio star flash transmission noise management method suitable for a conference system is characterized in that: The step of acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and performing noise reduction management on the ambient sound audio specifically includes: Acquire audio classification information, extract ambient sound audio information according to the audio classification information, form an ambient sound audio sequence based on the time sequence attribute of the ambient sound audio, and extract human voice audio information through the audio classification information to form a human voice audio sequence; Based on the ambient sound audio sequence, an average spectrum of the noise sequence is calculated as the noise spectrum, the ambient sound audio sequence is time-aligned with the human voice audio sequence, and a sequence where the human voice and the ambient sound overlap is marked to obtain a marked sequence; Extracting features of the ambient sound audio information to obtain frequency, amplitude and audio waveform features, and performing periodic analysis to obtain periodic analysis information; The spectrum of the noisy frequency is calculated according to the marker sequence, and subtracted from the noise spectrum to obtain the subtracted spectrum. The correction parameters are generated through the periodic analysis information, and the subtracted spectrum is corrected to obtain the corrected spectrum. The inverse Fourier transform is used to generate the ambient sound audio noise reduction result information.

5. According to claim 1, a wireless audio star flash transmission noise management method suitable for a conference system is characterized in that: The method of acquiring human voice audio information according to the audio classification information, performing human voice audio analysis, dividing the human voice audio into background human voice audio and target human voice audio, performing noise reduction on the background human voice audio, and performing abnormality detection on the target human voice audio specifically includes: Acquire mixed human voice audio based on historical data retrieval, acquire a spectrogram of the mixed human voice audio according to the acquired mixed human voice audio, and divide the mixed human voice audio according to frequency to obtain first division information; Obtain the frequency and sound features of the background human voice and the target human voice in the speaking scene through instance statistical analysis, and divide the mixed human voice audio into the background human voice and the target human voice in combination with the first division information to form an instance data set; Acquire audio classification information, acquire human voice audio information according to the audio classification information, perform pre-emphasis filtering on the human voice audio information, and perform feature extraction to acquire human voice audio feature information; Constructing a human voice audio analysis model based on SVM, performing model training through the example data set to obtain a human voice audio analysis model that meets expectations, inputting the human voice audio feature information into the human voice audio analysis model for analysis, and obtaining human voice audio analysis information; The human voice audio analysis information divides the human voice audio into background human voice and target human voice, and uses spectrum subtraction to reduce background human voice noise; Extract target human voice audio based on human voice audio analysis information, build an anomaly detection model through generative adversarial network, input target human voice audio and generate original human voice audio sequence according to target human voice audio features, and obtain generated human voice audio sequence through generator; A reconstruction error between the original human voice audio sequence and the generated human voice audio sequence is calculated, and abnormal human voice audio detection is performed according to the reconstruction error to obtain abnormal human voice audio detection information.

6. According to a wireless audio star flash transmission noise management method applicable to a conference system according to claim 1, it is characterized in that: The comfort level of the target human voice audio is evaluated based on the abnormality detection result, whether there is artificial noise and audio prediction is performed, and noise reduction control parameters are generated to perform noise reduction management on the target human voice audio, specifically including: Based on data retrieval, the audio of the speaker's voice under each amplifier standard in different conference scenarios is obtained, and the comfort level of each speaker's voice audio is evaluated through expert analysis to obtain the audio of the speaker's voice with different comfort levels; Extracting human voice audio features corresponding to human voice audio with different comfort levels to construct comfort evaluation rules, obtaining abnormal human voice audio detection information, and performing comfort evaluation on abnormal human voice audio to obtain comfort evaluation information; If the comfort evaluation result is that the target human voice audio at the current moment is comfortable, then voice enhancement is performed on the target human voice audio at the current moment; If the comfort evaluation result is that the target human voice audio at the current moment is uncomfortable, then artificial noise appears in the target human voice audio at the current moment, noise management is performed, and abnormal human voice audio features are extracted according to the abnormal human voice audio detection information; Obtaining a comfortable audio benchmark according to the comfort evaluation rule, performing deviation calculation on the extracted abnormal human voice audio feature to obtain a deviation value, performing noise reduction control on the abnormal human voice audio based on the calculated deviation value, and generating a noise reduction control parameter; The Markov algorithm is introduced to perform audio prediction based on the extracted abnormal human voice audio features, the initial state is set according to the abnormal human voice audio features, and the state transfer matrix of the initial state is retrieved in the preset state space; Analyze the state transition probability of the next state of the abnormal human voice audio at the current moment according to the state transition matrix, and generate a state transition dynamic sequence of the abnormal human voice audio at the current moment; Acquire human voice audio analysis information, extract target human voice audio features and generate the actual state change sequence of the target human voice, analyze the state transition probability of the actual audio change, select the state transition sequence that conforms to the actual audio change in combination with the state transition dynamic sequence, and obtain the final state transition sequence; Generating an audio prediction result according to the final state transition sequence, performing anomaly detection and comfort evaluation on the audio prediction result, and performing noise reduction control on the audio prediction result whose evaluation result is uncomfortable, to generate a noise reduction control prediction parameter; The target human voice audio at the next moment is obtained for feature extraction and the audio state change sequence at the next moment is generated, which is compared and analyzed with the final state transition sequence, and the noise reduction control prediction parameters are corrected according to the comparison and analysis results.

7. A wireless audio star flash transmission noise management system suitable for conference systems, characterized in that: The system includes: a memory and a processor, wherein the memory contains a wireless audio star flash transmission noise management method program applicable to a conference system, and when the wireless audio star flash transmission noise management method program applicable to a conference system is executed by the processor, the following steps are implemented: Obtain the audio information to be transmitted of the target conference scene, encode the audio to be transmitted, and perform preliminary separation of the audio to be transmitted based on the non-negative matrix decomposition algorithm to obtain separated audio information; Constructing an audio classification model, performing audio classification according to the separated audio information, classifying the audio to be transmitted into different audio types, and obtaining audio classification information; Acquiring ambient sound audio information according to the audio classification information, forming an ambient sound audio sequence based on the time sequence attributes of the ambient sound audio, and performing noise reduction management on the ambient sound audio; Acquire human voice audio information according to the audio classification information, perform human voice audio analysis, divide the human voice audio into background human voice audio and target human voice audio, perform noise reduction on the background human voice audio, and perform anomaly detection on the target human voice audio; Based on the anomaly detection results, the comfort level of the target vocal audio is evaluated, whether there is artificial noise is determined and audio prediction is performed, and noise reduction control parameters are generated to manage the noise reduction of the target vocal audio.

Images