A method and system for noise reduction processing of communication audio data

Abstract

This invention belongs to the field of data noise reduction technology, specifically relating to a method and system for noise reduction processing of communication audio data. The method includes: when acquiring communication audio data to be processed and corresponding background audio data, performing noise reduction processing on the communication audio data to obtain denoised corrected data. This invention preprocesses the communication audio data by comparing sampling offset correction and sound pressure level difference. It determines the sampling offset by comparing the sound start time points of the communication audio data and the background audio data, and compares the sound pressure level difference between the corrected data and the background audio data. The difference parameters are classified into difference type parameters, and negative difference portions are then removed as noise data. This precise location and efficient removal of noise data in the communication audio data improves the accuracy of noise reduction processing and the purity of the audio data.

Description

Technical Field

[0001] This invention belongs to the field of data noise reduction technology, specifically relating to a method and system for noise reduction processing of communication audio data. Background Technology

[0002] In various communication scenarios, the clarity and integrity of audio signals determine the accuracy of information transmission and the efficiency of interaction. Ensuring that audio signals maintain fidelity in complex and ever-changing transmission environments guarantees the stable operation of communication systems and improves user experience.

[0003] Existing audio noise reduction technologies, when dealing with complex environments, rely on traditional methods such as filtering, waveform analysis, and spectral subtraction. These methods lack adaptability to dynamic noise environments. When faced with non-stationary noise or sudden interference, they cannot distinguish between the target signal and background noise, easily leading to over-processing of the audio signal and resulting in signal distortion or loss of crucial voice information. Furthermore, traditional processing flows often involve cumbersome computational steps, resulting in low processing efficiency, difficulty in meeting the low-latency requirements of real-time communication, and significant system overhead when deployed on resource-constrained terminal devices.

[0004] In view of this, the present invention proposes a method and system for noise reduction processing of communication audio data. Summary of the Invention

[0005] The purpose of this invention is to provide a method for noise reduction processing of communication audio data, which solves the problems of low processing efficiency and delay in the noise reduction processing of communication audio data.

[0006] The technical solution adopted by the present invention is as follows: a method for noise reduction processing of communication audio data, wherein when the communication audio data to be processed and the corresponding background audio data are obtained, the following steps are performed: noise reduction processing is performed on the communication audio data to obtain the noise-reduced correction data; Based on the denoised correction data, determine the target optimization strategy; Among them, the noise reduction processing of communication audio data to obtain noise-reduced correction data includes: determining the noise distribution of communication audio data based on the sound pressure value distribution of communication audio data and background audio data; The starting time points of the communication audio data and the background audio data are compared to determine the sampling offset; the communication audio data is then corrected based on the sampling offset to obtain the corrected data. And compare the correction data with the sound pressure value of the background audio data, and based on the comparison results, identify and remove the noise data in the communication audio data; The process of comparing the start time points of the communication audio data and the background audio data to determine the sampling offset includes: comparing the sampling offset with a set threshold to determine whether the communication audio data needs correction; if the communication audio data needs correction, generating the sampling offset; if the communication audio data does not need correction, directly outputting the communication audio data as correction data. The comparison of sound pressure levels between the correction data and the background audio data includes: comparing the sound pressure level differences between the correction data and the background audio data; generating difference parameters for the correction data; performing segmentation processing on the difference parameters; and classifying the difference parameters into difference type parameters.

[0007] Preferably, the negative difference portion of the difference type parameter is discarded as noise data, while the positive difference portion is retained and determined as valid data.

[0008] Preferably, the target optimization strategy is determined based on the denoised correction data, including: performing information sampling based on the denoised correction data and extracting key nodes and ordinary nodes from the sampling results; and generating simulation scoring parameters based on the key nodes and ordinary nodes. The denoised correction data is processed using an optimization strategy; based on the simulated scoring parameters, the score value corresponding to the optimization strategy is matched, and the highest score value is marked as the quality score; and based on the quality score, the corresponding optimization strategy is determined as the target optimization strategy.

[0009] Preferably, based on key nodes and ordinary nodes, simulated scoring parameters are generated, including: performing simulated processing by applying a preset optimization strategy to the denoised correction data, and extracting nodes from the simulated processing results; Identify key nodes among the nodes; calculate the proportion of key nodes under the preset optimization strategy; and generate simulated scoring parameters based on the proportion of key nodes.

[0010] Preferably, matching the score value corresponding to the optimization strategy includes: assigning a score based on a preset weight during the calculation of the quality score; and calculating the comprehensive score under the optimization strategy by combining the simulated scoring parameters, comparing the comprehensive score, and determining the optimization strategy corresponding to the highest score value as the target optimization strategy.

[0011] Preferably, the method further includes: generating data processing results of the target optimization strategy after the target optimization strategy is output; and comparing the data processing results with quality control parameters. If the data processing result is better than or equal to the quality control parameter, the official result is output; and if the data processing result is lower than the quality control parameter, a backup optimization strategy is output.

[0012] Preferably, the difference parameter is segmented, including dividing the difference parameter into intervals according to a preset threshold.

[0013] Preferably, the preset weights are pre-set based on the degree of influence of different simulated scoring parameters on the final quality score.

[0014] Preferably, the quality control parameter is a threshold used to evaluate whether the data processing results meet the preset quality standards.

[0015] A communication audio data noise reduction processing system, comprising: The data acquisition module is used to acquire the communication audio data to be processed and the corresponding background audio data; The data correction module is used to respond to the output of the data acquisition module, compare the sound start time points of the communication audio data and the background audio data to determine the sampling offset, and correct the communication audio data according to the sampling offset to obtain the corrected data. The noise removal module responds to the output of the data correction module by comparing the correction data with the sound pressure value of the background audio data. Based on the comparison result, it identifies and removes the noise data in the communication audio data to obtain the noise-reduced correction data. The scoring generation module responds to the output of the noise removal module, performs information sampling based on the denoised correction data, and extracts key nodes and ordinary nodes from the sampling results to generate simulated scoring parameters. The strategy determination module responds to the output of the scoring generation module, matches the score value corresponding to the optimization strategy based on the simulated scoring parameters, and marks the highest score value as the quality score in order to determine the target optimization strategy. The system also includes a results evaluation module, which responds to the output of the strategy determination module, generates data processing results for the target optimization strategy, and compares the data processing results with quality control parameters to output formal results or alternative optimization strategies based on the comparison results.

[0016] Beneficial effects: 1. This invention preprocesses communication audio data by comparing the sampling offset correction and sound pressure value difference. It determines the sampling offset by comparing the sound start time points of the communication audio data and the background audio data, and compares the sound pressure value difference between the corrected data and the background audio data. The difference parameter is divided into difference type parameters, and then the negative difference part is removed as noise data. This allows for the accurate positioning and efficient removal of noise data in communication audio data, improving the accuracy of noise reduction processing and the purity of audio data.

[0017] 2. This invention employs a dynamic strategy optimization based on simulated scoring parameters. By sampling the denoised correction data, key nodes and ordinary nodes are extracted. The proportion of key nodes and delay offset are estimated based on historical performance parameters or complexity models. Combined with preset weights, a comprehensive score is calculated, and the target optimization strategy corresponding to the highest score value is selected from multiple optimization strategies. Based on the real-time characteristics of the audio data, the target optimization strategy is automatically matched and executed, solving the problem of difficulty in fixing denoising parameters in existing technologies.

[0018] 3. After the target optimization strategy is output, the present invention compares the data processing results with the quality control parameters in real time. If the data processing results do not meet the standards, it automatically switches to the backup optimization strategy for processing, thereby avoiding the risk of a single strategy failure and ensuring that the final output audio data meets the expected quality standards. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Please refer to Figure 1 This embodiment provides a method for noise reduction processing of communication audio data, including the following steps: S1. Obtain the parameter distribution map and reflect the difference in the distribution of noise and effective signal by statistically analyzing the frequency or probability density of different sound pressure levels. Specifically, obtain several communication audio data and their corresponding background audio data, where the communication audio data contains effective speech signals and noise signals, while the background audio data contains only noise signals. All of these data were collected from the same recording environment. The sound pressure distribution of the acquired communication audio data was analyzed to form a data structure that characterizes the statistical characteristics of the sound pressure values ​​of the communication audio data within a specific time window. By statistically analyzing the frequency or probability density of different sound pressure levels, the distribution differences between noise and effective signals can be reflected, and the noise distribution in communication audio data can be identified. By identifying discontinuous bands with short durations and drastic intensity changes in sound pressure values, such as brief background noise, instantaneous impact sounds, or discontinuous human voice interference, the noise distribution characteristics in communication audio data can be determined.

[0022] To verify the accuracy of the communication audio data and background audio data in subsequent steps, the communication audio data is designated as communication audio data, and the background audio data is designated as background audio data.

[0023] S2. Sampling verification: By comparing the start time points of the communication audio data and the background audio data, the time deviation is calculated and aligned to ensure the data synchronization of subsequent noise reduction processing. Specifically, the sound start time point is detected for both communication audio data and background audio data. This detection process identifies the time point when the signal strength first continuously exceeds the preset activation threshold by analyzing the energy envelope or short-time average amplitude of the audio signal, and determines it as the sound start time point.

[0024] Then, the time difference between the two sound start times is calculated, and this difference is used as the sampling offset to measure the degree of asynchrony of the data on the time axis.

[0025] Then compare the sampling offset with the preset threshold: If the sampling offset is less than or equal to the set threshold, it is determined that the communication audio data and the background audio data are well aligned in time and no correction is required. In this case, the communication audio data is directly output as the correction data. If the sampling offset is greater than the set threshold, it is determined that there is a significant time deviation between the communication audio data and the background audio data, and correction is required. The system will shift the communication audio data along the time axis according to the calculated sampling offset to align it with the background audio data in time, thereby obtaining the correction data. If the sampling offset significantly exceeds the set threshold, it indicates that the communication audio data and the background audio data may not come from the same recording environment or there may be a serious acquisition problem. At this point, the system will mark the current data as invalid and terminate subsequent noise reduction processing, or trigger a data re-acquisition command.

[0026] S3. Perform noise removal by comparing the sound pressure values ​​of the correction data with the background audio data frame by frame or sample by sample. By calculating the instantaneous sound pressure difference between the two, a difference parameter is generated to quantify the noise components, which facilitates subsequent noise identification and removal.

[0027] Then, the difference parameters are segmented. For example, the range is divided according to a preset sound pressure difference range or time window. The sound pressure difference range is a preset range of sound pressure differences, which is used to classify the difference parameters so as to carry out differentiated processing for noise of different intensities. The difference parameters are divided into difference type parameters, such as high intensity negative difference, medium intensity negative difference, low intensity negative difference, etc. The high-intensity negative difference is the range where the sound pressure level of the corrected data is significantly lower than that of the background audio data, representing high-intensity noise components; the medium-intensity negative difference is the range where the sound pressure level of the corrected data is moderately lower than that of the background audio data, representing medium-intensity noise components; and the low-intensity negative difference is the range where the sound pressure level of the corrected data is slightly lower than that of the background audio data, representing low-intensity noise components.

[0028] Among these difference type parameters, the system identifies the negative difference portion as noise data, that is, the portion where the sound pressure value of the corrected data is lower than the sound pressure value of the background audio data.

[0029] For these noise data, the system removes them by attenuating or zeroing the amplitude of the audio signal in the corresponding time period, and initially determines and retains the positive difference as valid data, that is, the part of the corrected data sound pressure value that is higher than the background audio data sound pressure value. Through the above noise reduction process, the system obtains the initial noise reduction correction data. This step aims to initially purify the communication audio data by directly comparing and subtracting background noise components.

[0030] S4. Generate simulated scoring parameters. Based on the correction data after preliminary noise reduction, the system performs information sampling, such as feature extraction of audio signals, such as spectrum analysis, energy envelope analysis, or instantaneous frequency analysis; and extracts key nodes and ordinary nodes from the sampling results. Key nodes represent key signal features or structural points in audio, such as the start point of speech, the stressed part of a syllable, or specific phoneme features, while ordinary nodes are other non-key signal points.

[0031] Based on these key nodes and ordinary nodes, the delay offset is output by comparing the time difference of specific events in the audio signal before and after noise reduction, such as the speech start point and syllable boundary.

[0032] For each preset optimization strategy, the system retrieves statistical performance parameters from its historical processing records for similar audio features such as signal-to-noise ratio, noise type, and energy envelope shape, or performs performance prediction based on the computational complexity model of the strategy, in order to obtain the proportion of key nodes corresponding to that strategy. And delay offset D; where, the proportion of critical nodes The ratio of the number of critical nodes to the total number of nodes is given, and the delay offset D is the time difference of a specific event before and after the noise reduction process. Each of the preset optimization strategies includes, but is not limited to, Wiener filtering-based denoising strategies, spectral subtraction-based denoising strategies, wavelet threshold shrinking-based denoising strategies, and lightweight neural network-based denoising strategies.

[0033] Then combine the proportion of this key node Delay offset D and total number of nodes The simulated scoring parameters S are generated through a preset scoring model, which is specifically defined as follows:

[0034] Algorithm explanation: A combination of weighted summation and exponential decay function is used, where... It is a natural exponential function. It is the natural logarithm function.

[0035] In the formula, This represents the simulation scoring parameters, which are quantitative indicators used to measure the overall performance of the optimization strategy. This indicates the percentage of key nodes, which is the ratio of the number of key nodes to the total number of nodes. This represents the delay offset, which is the time difference between a specific event before and after the noise reduction process. This represents the total number of nodes, which is the total number of all feature nodes extracted from the audio signal. This represents the weighting coefficient, which signifies the importance weight of each evaluation dimension. This represents the delay sensitivity coefficient, which is a parameter that adjusts the degree to which delay affects the score.

[0036] Among them, the weighting coefficient The following method was used to determine this: At least 100 sets of communication audio test samples labeled with subjective quality scores were obtained, based on the proportion of key nodes. Delay offset D, total number of nodes The input features are subjective quality ratings, with subjective quality ratings as the target variable. Multiple linear regression was used to fit the regression coefficients of each feature, and the normalized regression coefficients were then used as the regression coefficients. The value of λ is determined by a grid search method, which iterates through the interval [0.01, 1.0] with a step size of 0.01 to select the value of λ that maximizes the Pearson correlation coefficient between the rating and the subjective quality rating. This mechanism enables the noise reduction system to make intelligent decisions by optimizing strategies based on the real-time characteristics of audio data.

[0037] S5. Output optimization strategies. Based on the simulated scoring parameters generated in step S4, evaluate the preset optimization strategies. Specifically, according to the preset weights and combined with the simulated scoring parameters of each strategy, calculate the comprehensive score of each optimization strategy, which is used to measure the overall performance of the optimization strategy.

[0038] By comparing these comprehensive scores, the system determines the optimization strategy with the highest score as the target optimization strategy, which is used as the algorithm configuration scheme for performing noise reduction processing on communication audio data.

[0039] The system then applies the target optimization strategy to the corrected data after initial noise reduction, performs actual noise reduction processing, and generates the data processing result of the target optimization strategy. The system compares this data processing result with the quality control parameters to ensure the quality of the output audio data. If the quality index of the data processing result is greater than or equal to the quality control parameter, the system will output this data processing result as the official result. The quality index includes the signal-to-noise ratio (SNR) and speech intelligibility index. The quality control parameter is an independent SNR threshold or PESQ scoring threshold, which is used to objectively evaluate the audio quality after actual noise reduction processing. It belongs to a different evaluation dimension than the simulated scoring parameter.

[0040] If the quality index of the data processing result is lower than the quality control parameter, it indicates that the current target optimization strategy has failed to achieve the expected quality standard. At this point, the system will automatically select a backup optimization strategy, such as a general robust strategy or the strategy with the second highest score in the S4 evaluation, and apply it to the corrected data after the initial noise reduction for further noise reduction processing. Then, the data processing results generated by the alternative optimization strategy are output to ensure the quality of the output audio data and the robustness of the system.

[0041] Among them, the AI ​​strategy optimizes the evaluation logic of human experts, pre-sets a multi-dimensional feature scoring model based on factors such as the proportion of key nodes and delay offset, i.e., simulated scoring parameters, and uses this model to automatically and intelligently simulate and comprehensively score various candidate noise reduction strategies, and finally autonomously decides and executes the optimal strategy.

[0042] Example 2: Please refer to Figure 2 This embodiment provides a communication audio data noise reduction processing system, which matches efficient noise reduction and optimization strategies for communication audio data. The system includes the following modules: The data acquisition module is used to acquire the communication audio data to be processed and the corresponding background audio data.

[0043] The data correction module responds to the output of the data acquisition module by comparing the start time points of the communication audio data and the background audio data to determine the sampling offset, and then corrects the communication audio data based on the sampling offset to obtain corrected data.

[0044] Specifically, this module analyzes the energy envelope of the audio signal to identify the time point when the signal strength first continuously exceeds the preset activation threshold, calculates the difference between the start time point of the communication audio data and the start time point of the background audio data, and obtains the sampling offset. The sampling offset is compared with a set threshold. If the sampling offset is less than or equal to the set threshold, it is determined that no correction is needed, and the communication audio data is directly output as the correction data. If the sampling offset is greater than the set threshold, it is determined that correction is needed, and the communication audio data is shifted along the time axis according to the sampling offset to obtain the correction data. For example, when the preset activation threshold is -40dB and the set threshold is 5ms, if the communication audio start time is 100ms, the background audio start time is 102ms, and the sampling offset is 2ms, which is less than 5ms, then it is determined that no correction is needed; if the background audio start time is 110ms and the sampling offset is 10ms, which is greater than 5ms, then it is determined that correction is needed, and the system will shift the communication audio data backward by 10ms on the time axis.

[0045] The noise removal module responds to the output of the data correction module by comparing the correction data with the sound pressure level of the background audio data. Based on the comparison result, it identifies and removes the noise data in the communication audio data to obtain the noise-reduced correction data.

[0046] Specifically, this module calculates the difference between the sound pressure level of the correction data and the sound pressure level of the background audio data at the same sampling time point, and divides the difference parameter into intervals according to a preset threshold, classifying it into difference type parameters. If the difference parameter is negative, that is, the sound pressure level of the correction data is less than the sound pressure level of the background audio data, it is determined to be noise data and is discarded; if the difference parameter is positive, that is, the sound pressure level of the correction data is greater than the sound pressure level of the background audio data, it is determined to be valid data and is retained. For example, assuming the sound pressure level of the background audio is 50dB at a certain moment, if the sound pressure level of the correction data at that moment is 45dB, the difference is -5dB, which is determined to be noise, and the data at that moment is set to zero or attenuated; if the sound pressure level of the correction data at that moment is 55dB, the difference is +5dB, which is determined to be valid data and is retained.

[0047] The scoring generation module responds to the output of the noise removal module, performs information sampling based on the denoised correction data, and extracts key nodes and ordinary nodes from the sampling results to generate simulated scoring parameters.

[0048] Specifically, this module stores a historical processing performance database for each preset optimization strategy, or has a built-in computational complexity analysis model for each strategy. These preset optimization strategies include, but are not limited to, Wiener filtering-based denoising strategies, spectral subtraction-based denoising strategies, wavelet threshold shrinking-based denoising strategies, and lightweight neural network-based denoising strategies. For each preset optimization strategy, the module retrieves its historical processing performance parameters or estimates its theoretical performance based on a complexity model to obtain the key node ratio and delay offset corresponding to that strategy. The module calls a simulation scoring parameter function, calculating the simulation scoring parameters based on the key node ratio, delay offset, and total number of nodes using a weighted summation and exponential decay function. The weight coefficients are obtained through multiple linear regression fitting, and the delay sensitivity coefficient is obtained through grid search optimization.

[0049] Specifically, the module calculates the comprehensive score of each optimization strategy based on preset weights and simulated scoring parameters, compares the comprehensive scores, and selects the strategy with the highest score as the target optimization strategy. For example, if there are candidate optimization strategies A and B, strategy A has a comprehensive score of 1.505 and strategy B has a comprehensive score of 1.614, then strategy B is determined as the target optimization strategy.

[0050] The results evaluation module responds to the output of the strategy determination module, generates the data processing results of the target optimization strategy, and compares the data processing results with the quality control parameters to output the formal results or alternative optimization strategies based on the comparison results.

[0051] Specifically, this module applies a target optimization strategy to denoise the correction data, generates data processing results, and compares the data processing results with quality control parameters: If the data processing result is better than or equal to the quality control parameter, then the official result is output; If the data processing result is lower than the quality control parameter, an alternative optimization strategy is output and the noise reduction process is re-executed. For example, assuming the quality control parameter is 30dB. If the signal-to-noise ratio of the data after the target optimization strategy is 32dB, it is considered qualified and the official result is output. If the signal-to-noise ratio is 28dB, it is deemed unqualified, and the system automatically switches to the backup optimization strategy for processing.

[0052] The quality control parameters are signal-to-noise ratio thresholds or PESQ scoring thresholds, which are independent of the simulated scoring parameters. They are used to objectively evaluate the audio quality after actual noise reduction processing and belong to different evaluation dimensions than the simulated scoring parameters.

[0053] Through the collaboration of the above modules, a complete closed-loop processing flow is formed, from data acquisition, correction, noise removal, score generation, strategy determination to result evaluation. It is suitable for high-quality noise reduction processing of communication audio data in complex background noise environments. The specific implementation of the above modules is only a preferred embodiment of the present invention and does not constitute a limitation of the present invention.

Claims

1. A method for noise reduction processing of communication audio data, characterized in that, When the communication audio data to be processed and the corresponding background audio data are obtained, execute: The communication audio data is denoised to obtain the denoised correction data. Based on the denoised correction data, a target optimization strategy is determined; this includes denoising the communication audio data to obtain the denoised correction data, including: Based on the sound pressure level distribution of the communication audio data and the background audio data, the noise distribution of the communication audio data is determined; the sound start time points of the communication audio data and the background audio data are compared to determine the sampling offset. The communication audio data is corrected based on the sampling offset to obtain corrected data; and the corrected data is compared with the sound pressure value of the background audio data. Based on the comparison result, the noise data in the communication audio data is identified and removed. The process of comparing the start time points of the communication audio data and the background audio data to determine the sampling offset includes: comparing the sampling offset with a set threshold to determine whether the communication audio data needs correction; if the communication audio data needs correction, generating the sampling offset; if the communication audio data does not need correction, directly outputting the communication audio data as correction data. The comparison of sound pressure levels between the correction data and the background audio data includes: comparing the sound pressure level differences between the correction data and the background audio data; generating difference parameters for the correction data; performing segmentation processing on the difference parameters; and classifying the difference parameters into difference type parameters.

2. The method for noise reduction processing of communication audio data according to claim 1, characterized in that, Negative difference values ​​in the difference type parameter are discarded as noise data, while positive difference values ​​are retained and identified as valid data.

3. The method for noise reduction processing of communication audio data according to claim 1, characterized in that, Based on the denoised and corrected data, the target optimization strategy is determined, including: Based on the denoised and corrected data, information sampling is performed, and key nodes and ordinary nodes are extracted from the sampling results; For the various preset optimization strategies, the historical processing performance parameters of each strategy are retrieved or the performance is estimated based on the strategy complexity model to obtain the key node ratio and delay offset corresponding to each strategy; based on the key node ratio and delay offset, the simulated scoring parameters of each strategy are generated. The denoised correction data is processed using an optimization strategy; based on the simulated scoring parameters, the score value corresponding to the optimization strategy is matched, and the highest score value is marked as the quality score. And based on the quality score, the corresponding optimization strategy is determined as the target optimization strategy.

4. The method for noise reduction processing of communication audio data according to claim 3, characterized in that, The simulated scoring parameters are generated as follows: For each preset optimization strategy, retrieve the performance parameters for the same audio features from its historical processing records, or perform performance prediction based on the computational complexity model of the strategy to obtain the key node ratio and delay offset corresponding to the strategy; and generate simulated scoring parameters based on the key node ratio and delay offset through the preset scoring model.

5. The method for noise reduction processing of communication audio data according to claim 3, characterized in that, The score value corresponding to the matching optimization strategy includes: the score assigned based on preset weights during the calculation of the quality score; Furthermore, by combining the simulated scoring parameters, the comprehensive score under the optimization strategy is calculated, and by comparing the comprehensive scores, the optimization strategy corresponding to the highest score value is determined as the target optimization strategy.

6. The method for noise reduction processing of communication audio data according to claim 1, characterized in that, The method also includes: generating the data processing results of the target optimization strategy after the target optimization strategy is output; Compare the data processing results with quality control parameters; If the data processing result is better than or equal to the quality control parameter, the official result is output; and if the data processing result is lower than the quality control parameter, a backup optimization strategy is output.

7. The method for noise reduction processing of communication audio data according to claim 1, characterized in that, The difference parameter is segmented, including dividing the difference parameter into intervals according to a preset threshold.

8. The method for noise reduction processing of communication audio data according to claim 5, characterized in that, The preset weights are obtained by normalizing the regression coefficients obtained from multiple linear regression fitting of audio test samples labeled with subjective quality scores; the delay sensitivity coefficient is obtained by grid search optimization within a preset interval.

9. A method for noise reduction processing of communication audio data according to claim 6, characterized in that, Quality control parameters are thresholds used to evaluate whether the data processing results meet preset quality standards.

10. A noise reduction processing system for communication audio data, characterized in that, include: The data acquisition module is used to acquire the communication audio data to be processed and the corresponding background audio data; The data correction module is used to respond to the output of the data acquisition module, compare the sound start time points of the communication audio data and the background audio data to determine the sampling offset, and correct the communication audio data according to the sampling offset to obtain the corrected data. The noise removal module responds to the output of the data correction module by comparing the correction data with the sound pressure value of the background audio data. Based on the comparison result, it identifies and removes the noise data in the communication audio data to obtain the noise-reduced correction data. The scoring generation module responds to the output of the noise removal module, performs information sampling based on the denoised correction data, and extracts key nodes and ordinary nodes from the sampling results to generate simulated scoring parameters. The strategy determination module responds to the output of the scoring generation module, matches the score value corresponding to the optimization strategy based on the simulated scoring parameters, and marks the highest score value as the quality score in order to determine the target optimization strategy. The system also includes a results evaluation module, which responds to the output of the strategy determination module, generates data processing results for the target optimization strategy, and compares the data processing results with quality control parameters to output formal results or alternative optimization strategies based on the comparison results.

Images