Sound data processing method and device, computer equipment and readable storage medium
Through the interference processing, clustering and clustering enhancement processing of sound data in the financial sector, the problems of not accurate sound polycosty results in the existing technology have been solved, and higher identity certification accuracy and financial security are realized Essence
Patent Information
- Application Number
- CN202411364495.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-01-24
AI Technical Summary
When the existing voiceprint clustering method recognizes the speaker's identity in the financial field, the clustering results are not accurate enough, which affects the accuracy of identity authentication.
By obtaining the sound data to be processed, performing de-interference processing to remove noise and voice broadcasting sounds, and then performing clustering processing on high-quality sound data, extracting voiceprint features, and performing cluster enhancement processing based on the similarity of these features to improve Reliability of clustering results.
It improves the accuracy of voiceprint feature extraction and clustering, enhances the reliability of voice data clustering results, improves the accuracy of identifying and verifying speaker identity, provides a more reliable identity authentication solution for the financial field, and improves financial security.
Smart Images
Figure CN119360857A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of artificial intelligence and financial technology, and in particular to a method, device, computer equipment and readable storage medium for processing sound data. Background Art
[0002] In the financial field, voiceprint clustering technology is needed to segment and cluster the voices of different speakers in the audio signal in anti-fraud scenarios such as finding one person with multiple accounts or one account with multiple people, and in scenarios such as financial meetings where speakers need to be automatically separated. This allows the speaker to be identified through voiceprint clustering technology.
[0003] At present, the mainstream voiceprint clustering methods include methods based on statistical models, methods based on deep learning, and methods based on clustering algorithms, etc. However, these methods all have the problem that the clustering results are not accurate enough. Summary of the invention
[0004] In view of this, the present application provides a method, apparatus, computer device and readable storage medium for processing sound data, which solve the problem of inaccurate clustering results.
[0005] In a first aspect, an embodiment of the present application provides a method for processing sound data, including:
[0006] Acquire sound data to be processed, and perform interference removal processing on the sound data;
[0007] Clustering the sound data after the interference removal process to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features;
[0008] Based on the similarity of the voiceprint features in each of the sound data clusters, cluster enhancement processing is performed on the multiple sound data clusters to obtain a sound data clustering result.
[0009] In a second aspect, an embodiment of the present application provides a device for processing sound data, including:
[0010] A data acquisition module, used for acquiring sound data to be processed;
[0011] An interference removal processing module, used for performing interference removal processing on the sound data;
[0012] A clustering processing module, used for performing clustering processing on the sound data after the interference removal processing to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features;
[0013] The cluster enhancement processing module is used to perform cluster enhancement processing on the multiple sound data clusters based on the similarity of the voiceprint features in each sound data cluster to obtain a sound data clustering result.
[0014] In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method of the first aspect are implemented.
[0015] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the method of the first aspect are implemented.
[0016] In an embodiment of the present application, sound data to be processed is obtained, and a de-interference process is performed on the sound data; clustering is performed on the sound data after the de-interference process to obtain multiple sound data clusters, each sound data cluster includes multiple voiceprint features; based on the similarity of the voiceprint features in each sound data cluster, clustering enhancement is performed on the multiple sound data clusters to obtain sound data clustering results. In the present application, de-interference processing is performed on the sound data to be processed, and clustering processing is performed based on relatively high-quality sound data to ensure the accuracy of voiceprint feature extraction and clustering. In addition, clustering enhancement processing is performed after clustering, which can enhance the reliability of the sound data clustering results, improve the accuracy of identifying and verifying the speaker's identity, provide a more reliable identity authentication solution for the financial field, and improve financial security.
[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0019] Figure 1 A schematic diagram showing a flow chart of a method for processing sound data according to an embodiment of the present application;
[0020] Figure 2 A schematic diagram showing the discrete situation of the sound features of the sound data clustering according to an embodiment of the present application;
[0021] Figure 3 A schematic diagram showing the density of sound data clustering according to an embodiment of the present application;
[0022] Figure 4 A structural block diagram of a device for processing sound data according to an embodiment of the present application is shown;
[0023] Figure 5 One of the structural block diagrams of the computer device according to the embodiment of the present application is shown;
[0024] Figure 6 The second structural block diagram of the computer device according to the embodiment of the present application is shown. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] The following, in conjunction with the accompanying drawings, describes in detail the sound data processing method, apparatus, computer device, and readable storage medium provided in the embodiments of the present application through specific embodiments and their application scenarios.
[0027] The present application embodiment provides a method for processing sound data, such as Figure 1 As shown, the method includes:
[0028] Step 101, obtaining the sound data to be processed, and performing interference removal processing on the sound data;
[0029] Step 102, clustering the sound data after the interference removal process to obtain a plurality of sound data clusters, each of which includes a plurality of voiceprint features;
[0030] Step 103: Based on the similarity of the voiceprint features in each sound data cluster, cluster enhancement processing is performed on the multiple sound data clusters to obtain a sound data clustering result.
[0031] In the financial field, when users (such as consumers or investors) make inquiries or complaints by phone or online voice, or in offline or online meetings, the user's voice data is collected and clustered for rapid identification and verification of the speaker's identity.
[0032] In the embodiment of the present application, after the execution subject obtains the sound data to be processed, it first performs interference removal processing on the sound data to be processed, for example, removing background noise, echo, reverberation and other noises in the sound data, and removing the voice broadcast sound in the sound data. After the interference removal processing, relatively high-quality sound data is obtained, and then the sound data is clustered, that is, similar sound data is classified into one category, so that the sound data is effectively grouped.
[0033] In one embodiment, clustering the sound data includes: extracting stable and unique voiceprint features from the sound data, the voiceprint features can reflect the unique voiceprint characteristics of the speaker, including the frequency, timbre, speaking speed, intonation, etc. Selecting a suitable similarity metric (such as Euclidean distance, cosine similarity, etc.), calculating the similarity of voiceprint features between different sound segments, and using a clustering algorithm (such as a hierarchical clustering algorithm, a K-means clustering algorithm, etc.) to merge sound segments with higher similarity into one category until the preset clustering conditions are met.
[0034] In order to improve the accuracy of clustering results, the embodiment of the present application adds post-processing after clustering, that is, cluster enhancement processing. Based on the similarity of the voiceprint features in each sound data cluster, cluster enhancement processing is performed on multiple sound data clusters to enhance the reliability of the sound data clustering results. The sound data clustering results refer to the multiple sound data clusters finally obtained after the cluster enhancement processing.
[0035] Exemplarily, in the scenario of "one number for multiple people", multiple users share the same financial application account, which needs to be identified through voiceprint clustering analysis. Through the above-mentioned sound data processing method of the embodiment of the present application, after interference removal, the voiceprint features of multiple users are extracted, the sound data is clustered, and then cluster enhancement processing is performed to obtain the final sound data clustering result, that is, multiple sound data clusters. If it is found that the voiceprint features associated with a certain account are clustered into multiple different clusters, and the similarity between these clusters is low, it can be determined that the account has a "one number for multiple people" situation.
[0036] In the embodiment of the present application, the sound data to be processed is subjected to interference removal processing, and clustering processing is performed based on relatively high-quality sound data to ensure the accuracy of voiceprint feature extraction and clustering. In addition, clustering enhancement processing is performed after clustering, which can enhance the reliability of the sound data clustering results, improve the accuracy of identifying and verifying the speaker's identity, provide a more reliable identity authentication solution for the financial field, and improve financial security.
[0037] In one embodiment of the present application, the sound data is subjected to interference removal processing, including:
[0038] Detecting noise fragments in sound data;
[0039] If the sound data contains noise segments, the noise segments are removed to obtain non-noise sound data;
[0040] Detecting speech announcement segments in non-noise sound data;
[0041] If the non-noise sound data includes a voice broadcast segment, the voice broadcast segment in the non-noise sound data is removed to obtain the sound data after interference removal processing.
[0042] In this embodiment, the noise segment in the sound data can be detected by using a pre-built noise recognition model. When the sound data contains a noise segment, the noise segment is removed to retain the non-noise sound data, that is, retain the human voice data.
[0043] In one embodiment, the model can be trained according to the following method to obtain a sound recognition model:
[0044] 1) Collect sound datasets containing various types of noise in advance. The sound datasets include clean speech (i.e., only human voice) and speech with noise. The noise may include background noise, echo, reverberation, etc. This ensures the diversity and representativeness of the dataset to cover different environmental conditions and noise levels.
[0045] 2) Annotate the sound dataset to distinguish between human voice and noise, and to clarify whether each sample contains noise, the type of noise, and the intensity of the noise. These annotations will be used for training and supervised learning.
[0046] 3) Use time-frequency representation methods (such as Mel Frequency Cepstral Coefficients (MFCC)) and short-time Fourier transform (STFT) methods to extract features from sound signals. These features can effectively capture the key characteristics of sound, including frequency, energy distribution, and dynamic changes.
[0047] 4) Choose a long short-term memory network (LSTM), a convolutional neural network (CNN), or a combination of the two. These models usually perform well for noise recognition tasks and can effectively capture the temporal dependency and spatial characteristics of sound signals.
[0048] 5) Use the labeled dataset to train the selected model. During the training process, the model parameters, optimizer, and hyperparameters such as learning rate are adjusted to minimize the loss function (such as cross entropy loss), thereby improving the classification accuracy and generalization ability of the model.
[0049] The noise recognition model trained by the above method can accurately detect whether there are noise segments in the sound data.
[0050] Furthermore, for the non-noise sound data, it is continuously detected whether it contains a voice broadcast segment. When it is detected that the non-noise sound data contains a voice broadcast segment, the voice broadcast segment is removed to obtain the sound data after interference removal processing.
[0051] In one embodiment of the present application, detecting a voice announcement segment in non-noise sound data includes:
[0052] Compare the similarity between the voiceprint segment of the non-noise sound data and the sample speech broadcast segment in the sample voiceprint library;
[0053] If there is a sample voice broadcast segment in the sample voiceprint library whose similarity with the voiceprint segment of the non-noise sound data is greater than a first threshold, it is determined that the non-noise sound data contains the voice broadcast segment.
[0054] In this embodiment, the method of hitting the sample voiceprint library can be used to detect the voice broadcast segment. That is, the voice broadcast sounds encountered in financial scenarios are collected in advance, and the voiceprint features of the voice broadcast sounds are extracted through the voiceprint model to establish a sample voiceprint library, which stores various sample voice broadcast segments.
[0055] After obtaining the non-noise sound data, the voiceprint segment of the non-noise sound data is compared with the sample voice broadcast segment in the sample voiceprint library for voiceprint similarity. If there is a sample voice broadcast segment in the sample voiceprint library whose similarity with the voiceprint segment of the non-noise sound data is greater than the first threshold, that is, there is a sample voice broadcast segment that matches the voiceprint segment of the non-noise sound data, it means that the non-noise sound data contains a voice broadcast segment, and the segment is removed. After the voice broadcast is excluded, the remaining sound data will be a living human voice segment.
[0056] In the embodiment of the present application, after eliminating noise and voice broadcast sounds in turn, the low-quality data that is easy to cause interference is eliminated, and the remaining sound data will be high-quality living human voice fragments, which can be used to extract the voiceprint features of the speaker, avoiding the influence of noise and voice broadcast sounds on the human voice, improving the accuracy of voiceprint feature extraction, and thereby improving the accuracy of clustering.
[0057] In one embodiment of the present application, cluster enhancement processing is performed on multiple sound data clusters based on the similarity of the voiceprint features in each sound data cluster to obtain a sound data clustering result, including:
[0058] For each sound data cluster, discrete point exclusion processing and / or density filtering processing is performed based on the similarity of the voiceprint features in the sound data cluster to obtain the sound data clustering result; wherein, the clustering enhancement processing includes discrete point exclusion processing and / or density filtering processing, the discrete point exclusion processing is used to exclude discrete voiceprint features in the sound data cluster, and the density filtering processing is used to filter out sound data clusters with loose density.
[0059] In this embodiment, the cluster enhancement process includes a discrete point exclusion process and / or a density filtering process. For each voiceprint feature in each sound data cluster, a discrete point exclusion process is performed, and a density filtering process is performed on each sound data cluster.
[0060] The more discrete voiceprint features in the sound data clustering can be excluded by the discrete point exclusion process. For example, if the speaker corresponding to the sound data clustering is A, the more discrete voiceprint features in the sound data clustering refer to the features that are less likely to be determined as speaker A. Since voiceprint clustering is to obtain the voice segments of the same speaker, but does not require completeness, some segments that are not so certain to be the speaker can be selectively discarded, so as to select the segments, ensure the quality of the voiceprint, and improve the reliability of the clustering results. In addition, the discrete point exclusion process is a pruning method that can reduce the amount of data and ensure the timeliness of data processing.
[0061] Density filtering can filter out loose sound data clusters while retaining high-density sound data clusters. Loose sound data clusters can be understood as inaccurate clusters. Since voiceprint clustering is to obtain vocal segments of the same speaker, the application scenario of this application is mainly to obtain the target speaker's speaking segments, rather than to accurately score everyone's vocal segments. Therefore, some loose clusters can be appropriately abandoned to ensure the quality of voiceprints and improve the reliability of clustering results. In addition, density filtering can also reduce the amount of data and ensure the timeliness of data processing.
[0062] It should be noted that when the cluster enhancement process includes the discrete point exclusion process and the density filtering process, the density filtering process can be performed after the discrete point exclusion process, that is, it can be performed on the premise that the discrete points within the cluster are excluded.
[0063] The embodiment of the present application performs clustering enhancement processing after clustering, which can enhance the reliability of the sound data clustering results, improve the accuracy of identifying and verifying the speaker's identity, provide a more reliable identity authentication solution for the financial field, and improve financial security.
[0064] In one embodiment of the present application, based on the similarity of the voiceprint features in the sound data clusters, the sound data clusters are subjected to discrete point exclusion processing, including:
[0065] For any target voiceprint feature in the sound data clustering, calculate the similarity between the target voiceprint feature and other voiceprint features to obtain multiple similarities corresponding to the target voiceprint feature;
[0066] Calculate the average of multiple similarities as the discrete degree quantification value of the target voiceprint feature;
[0067] If the discrete degree quantization value is greater than or equal to the second threshold, the target voiceprint feature is retained in the sound data cluster; if the discrete degree quantization value is less than the second threshold, the target voiceprint feature is excluded from the sound data cluster.
[0068] In this embodiment, the discrete voiceprint features in each sound data cluster are excluded. Specifically, for each sound data cluster, each voiceprint feature in the cluster is traversed (the target voiceprint feature is any voiceprint feature in the cluster), and the similarity between it and the remaining voiceprint features is calculated to obtain multiple similarities, and then the arithmetic mean of the multiple similarities is calculated as the discrete degree quantization value of this voiceprint feature, indicating the discrete degree of this voiceprint feature.
[0069] The discrete degree quantization value is compared with the preset second threshold value. The voiceprint feature whose discrete degree quantization value is lower than the second threshold value is a discrete voiceprint feature. Since the discrete voiceprint feature is outlier, it is excluded. Figure 2 As shown, in the sound data clustering, voiceprint feature B is a discrete voiceprint feature, and voiceprint feature B needs to be excluded.
[0070] This method can achieve optimal segment selection, ensure voiceprint quality, and improve the reliability of clustering results. In addition, it can reduce the amount of data and ensure the timeliness of data processing.
[0071] In one embodiment of the present application, based on the similarity of the voiceprint features in the sound data clusters, density filtering is performed on the sound data clusters, including:
[0072] Based on the similarity of the voiceprint features in the sound data clusters, the density value of each sound data cluster is calculated;
[0073] In descending order of density values, a first preset number of sound data clusters are retained, and other sound data clusters beyond the first preset number are filtered out.
[0074] In this embodiment, for multiple sound data clusters, the density value of each sound data cluster is calculated, and the density value represents the degree of looseness of the sound features within the sound data cluster, such as Figure 3As shown, the density value of sound data cluster C is greater than the density value of sound data cluster D. Then, according to the scene situation or configuration setting, the first preset number is determined, and the first preset number of sound data clusters with larger density values are retained, and the rest will be filtered out to avoid affecting the voiceprint quality and improve the reliability of the clustering results. It can also reduce the amount of data and ensure the timeliness of data processing.
[0075] In one embodiment of the present application, based on the similarity of the voiceprint features in the sound data clusters, the density value of each sound data cluster is calculated, including:
[0076] For any sound data cluster, calculate the similarity between any two voiceprint features in the sound data cluster;
[0077] A second preset number of similarities are retained in descending order of similarity, and a standard deviation of the retained second preset number of similarities is calculated as a density value of the sound data clustering.
[0078] In this embodiment, the density value can be calculated using the half density method. Specifically, for each sound data cluster, the similarity between the multiple voiceprint features in the sound data cluster is calculated, and then the similarities are sorted from large to small, and the first second preset number of similarities are retained, for example, the first half of the high similarities are retained, and the standard deviation of the retained similarities is calculated as the density value of the sound data cluster. In this way, the density value of the sound data cluster can be accurately obtained, indicating the looseness of the sound features within the sound data cluster, thereby providing a data basis for cluster enhancement processing.
[0079] As a specific implementation of the above-mentioned sound data processing method, the present application embodiment provides a sound data processing device. Figure 4 As shown, the sound data processing device 400 includes: a data acquisition module 401, an interference removal processing module 402, a clustering processing module 403 and a clustering enhancement processing module 404.
[0080] The data acquisition module 401 is used to acquire the sound data to be processed;
[0081] The interference removal processing module 402 is used to perform interference removal processing on the sound data;
[0082] The cluster processing module 403 is used to perform cluster processing on the sound data after the interference removal processing to obtain multiple sound data clusters, each sound data cluster includes multiple voiceprint features;
[0083] The cluster enhancement processing module 404 is used to perform cluster enhancement processing on multiple sound data clusters based on the similarity of the voiceprint features in each sound data cluster to obtain a sound data clustering result.
[0084] In this embodiment, the sound data to be processed is subjected to interference removal, and clustering is performed based on relatively high-quality sound data to ensure the accuracy of voiceprint feature extraction and clustering. In addition, clustering enhancement is performed after clustering to enhance the reliability of the sound data clustering results, improve the accuracy of identifying and verifying the speaker's identity, provide a more reliable identity authentication solution for the financial field, and improve financial security.
[0085] Furthermore, the interference removal processing module 402 is specifically configured to:
[0086] Detecting noise segments in the sound data;
[0087] If the sound data contains noise segments, removing the noise segments to obtain non-noise sound data;
[0088] Detecting a speech broadcast segment in the non-noise sound data;
[0089] If the non-noise sound data includes a voice broadcast segment, the voice broadcast segment in the non-noise sound data is removed to obtain the sound data after interference removal processing.
[0090] Furthermore, the interference removal processing module 402 is specifically configured to:
[0091] Comparing the similarity between the voiceprint segment of the non-noise sound data and the sample speech broadcast segment in the sample voiceprint library;
[0092] If there is a sample voice broadcast segment in the sample voiceprint library whose similarity with the voiceprint segment of the non-noise sound data is greater than a first threshold, it is determined that the non-noise sound data contains a voice broadcast segment.
[0093] Furthermore, the clustering enhancement processing module 404 is specifically used to:
[0094] For each of the sound data clusters, discrete point exclusion processing and / or density filtering processing are performed based on the similarity of the voiceprint features in the sound data clusters to obtain the sound data clustering result; wherein the clustering enhancement processing includes the discrete point exclusion processing and / or the density filtering processing, the discrete point exclusion processing is used to exclude discrete voiceprint features in the sound data clusters, and the density filtering processing is used to filter out the sound data clusters with loose density.
[0095] Furthermore, the clustering enhancement processing module 404 is specifically used to:
[0096] For any target voiceprint feature in the sound data cluster, calculating the similarity between the target voiceprint feature and other voiceprint features to obtain multiple similarities corresponding to the target voiceprint feature;
[0097] Calculating an average value of a plurality of similarities as a quantized value of the discrete degree of the target voiceprint feature;
[0098] If the discrete degree quantization value is greater than or equal to a second threshold, the target voiceprint feature is retained in the sound data cluster; if the discrete degree quantization value is less than the second threshold, the target voiceprint feature is excluded from the sound data cluster.
[0099] Furthermore, the clustering enhancement processing module 404 is specifically used to:
[0100] Calculating density values of each of the sound data clusters based on similarities of the voiceprint features in the sound data clusters;
[0101] According to the order of the density values from large to small, a first preset number of the sound data clusters are retained, and other sound data clusters beyond the first preset number are filtered out.
[0102] Furthermore, the clustering enhancement processing module 404 is specifically used to:
[0103] For any of the voice data clusters, calculating the similarity between any two voiceprint features in the voice data cluster;
[0104] A second preset number of similarities are retained in descending order of the similarities, and a standard deviation of the second preset number of similarities retained is calculated as a density value of the sound data clustering.
[0105] The sound data processing device 400 in the embodiment of the present application can be a computer device, or a component in a computer device, such as an integrated circuit or a chip. The computer device can be a terminal, or other devices other than a terminal. The sound data processing device 400 provided in the embodiment of the present application can realize Figure 1 To avoid repetition, the various processes implemented in the embodiment of the method for processing sound data are not described here.
[0106] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, the functions or steps on the server side of a method for processing sound data are implemented.
[0107] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, the functions or steps on the client side of a method for processing sound data are implemented.
[0108] The embodiment of the present application further provides a computer device, which includes a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the following steps are implemented:
[0109] Acquire sound data to be processed, and perform interference removal processing on the sound data;
[0110] Clustering the sound data after the interference removal process to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features;
[0111] Based on the similarity of the voiceprint features in each of the sound data clusters, cluster enhancement processing is performed on the multiple sound data clusters to obtain a sound data clustering result.
[0112] The embodiment of the present application further provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the following steps are implemented:
[0113] Acquire sound data to be processed, and perform interference removal processing on the sound data;
[0114] Clustering the sound data after the interference removal process to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features;
[0115] Based on the similarity of the voiceprint features in each of the sound data clusters, cluster enhancement processing is performed on the multiple sound data clusters to obtain a sound data clustering result.
[0116] It should be noted that the above functions or steps that can be implemented by the computer device or readable storage medium can be found in the aforementioned method embodiments. To avoid repetition, they will not be described one by one here.
[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0118] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0119] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.
[0120] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A method for processing sound data, characterized in that: include: Acquire sound data to be processed, and perform interference removal processing on the sound data; Clustering the sound data after the interference removal process to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features; Based on the similarity of the voiceprint features in each of the sound data clusters, cluster enhancement processing is performed on the multiple sound data clusters to obtain a sound data clustering result.
2. The method according to claim 1, characterized in that The performing interference removal processing on the sound data includes: Detecting noise segments in the sound data; If the sound data contains noise segments, removing the noise segments to obtain non-noise sound data; Detecting a speech broadcast segment in the non-noise sound data; If the non-noise sound data includes a voice broadcast segment, the voice broadcast segment in the non-noise sound data is removed to obtain the sound data after interference removal processing.
3. The method according to claim 2, characterized in that The detecting of the voice broadcast segment in the non-noise sound data comprises: Comparing the similarity between the voiceprint segment of the non-noise sound data and the sample speech broadcast segment in the sample voiceprint library; If there is a sample voice broadcast segment in the sample voiceprint library whose similarity with the voiceprint segment of the non-noise sound data is greater than a first threshold, it is determined that the non-noise sound data contains a voice broadcast segment.
4. The method according to claim 1, characterized in that: The step of performing cluster enhancement processing on the plurality of sound data clusters based on the similarity of the voiceprint features in each sound data cluster to obtain a sound data clustering result includes: For each of the sound data clusters, discrete point exclusion processing and / or density filtering processing are performed based on the similarity of the voiceprint features in the sound data clusters to obtain the sound data clustering result; wherein the clustering enhancement processing includes the discrete point exclusion processing and / or the density filtering processing, the discrete point exclusion processing is used to exclude discrete voiceprint features in the sound data clusters, and the density filtering processing is used to filter out the sound data clusters with loose density.
5. The method according to claim 4, characterized in that Based on the similarity of the voiceprint features in the voice data cluster, the discrete point exclusion process is performed on the voice data cluster, including: For any target voiceprint feature in the sound data cluster, calculating the similarity between the target voiceprint feature and other voiceprint features to obtain multiple similarities corresponding to the target voiceprint feature; Calculating an average value of a plurality of similarities as a quantized value of the discrete degree of the target voiceprint feature; If the discrete degree quantization value is greater than or equal to a second threshold, the target voiceprint feature is retained in the sound data cluster; if the discrete degree quantization value is less than the second threshold, the target voiceprint feature is excluded from the sound data cluster.
6. The method according to claim 4, characterized in that Based on the similarity of the voiceprint features in the voice data cluster, the density filtering process is performed on the voice data cluster, including: Calculating density values of each of the sound data clusters based on similarities of the voiceprint features in the sound data clusters; According to the order of the density values from large to small, a first preset number of the sound data clusters are retained, and other sound data clusters beyond the first preset number are filtered out.
7. The method according to claim 6, characterized in that The calculating the density value of each of the sound data clusters based on the similarity of the voiceprint features in the sound data clusters includes: For any of the voice data clusters, calculating the similarity between any two voiceprint features in the voice data cluster; A second preset number of similarities are retained in descending order of the similarities, and a standard deviation of the second preset number of similarities retained is calculated as a density value of the sound data clustering.
8. A sound data processing device, characterized in that: include: A data acquisition module, used for acquiring sound data to be processed; An interference removal processing module, used for performing interference removal processing on the sound data; A clustering processing module, used for performing clustering processing on the sound data after the interference removal processing to obtain a plurality of sound data clusters, each sound data cluster including a plurality of voiceprint features; The cluster enhancement processing module is used to perform cluster enhancement processing on the multiple sound data clusters based on the similarity of the voiceprint features in each sound data cluster to obtain a sound data clustering result.
9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for processing sound data as described in any one of claims 1 to 7 when executing the computer program.
10. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps of the method for processing sound data according to any one of claims 1 to 7 are implemented.