Customer service voice recognition methods, devices, equipment and storage media
By filtering and clustering voice conversation information, separating customer service and visitor voice sequences, and performing multiple decoding operations, the problem of recognition accuracy in synthesized voice customer service systems has been solved, thus improving the accuracy of voice recognition.
Patent Information
- Application Number
- CN202210926324.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-03
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-08-03
AI Technical Summary
Existing intelligent voice customer service systems have poor accuracy in recognizing synthesized voices, resulting in low speaker separation and decoding accuracy, which cannot meet the needs of customer service scenarios.
By filtering the synthesized audio in the speech conversation to be recognized, clustering is performed to separate the voice sequences of customer service and visitors. After the initial decoding of these sequences, they are re-scored and decoded a second time to improve the recognition accuracy.
It effectively recognizes synthesized speech and improves the accuracy of speaker separation, thereby enhancing the speech recognition accuracy in customer service scenarios.
Smart Images

Figure CN117558273B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of speech recognition, specifically to a customer service speech recognition method, apparatus, device, and storage medium. Background Technology
[0002] Currently, with the continuous development of speech recognition technology, many industries have developed intelligent voice customer service systems based on this technology. These systems convert speech to text using speech recognition and then analyze the text using natural language processing algorithms to extract key information from customer service conversations. However, existing speech recognition systems use endpoint detection and speaker separation to segment speech into sequences before recognizing them, resulting in the loss of contextual information and inaccurate speech recognition. Secondly, customer service voices are often synthesized, and existing intelligent voice customer service systems do not consider synthesized voices during operation, leading to reduced accuracy in speaker separation and decoding, making it difficult to meet the needs of current customer service scenarios. Summary of the Invention
[0003] This application provides a customer service voice recognition method, apparatus, device, and storage medium, aiming to solve the technical problem of poor accuracy in voice recognition and speaker separation in existing intelligent voice customer service systems.
[0004] On one hand, this application provides a customer service voice recognition method, which includes the following steps:
[0005] Synthetic audio is filtered from the voice conversation information to be identified, and the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0006] The customer service voice sequence is initially decoded to obtain the customer service decoding result, and the visitor voice sequence is initially decoded to obtain the visitor decoding result.
[0007] The customer service voice sequence is re-scored based on the customer service decoding result, and the visitor voice sequence is re-scored based on the visitor decoding result;
[0008] The customer service voice sequence after rescoring is decoded a second time, and the visitor voice sequence after rescoring is decoded a second time to obtain the voice recognition result of the voice conversation information.
[0009] In one possible implementation of this application, the initial decoding of the customer service voice sequence to obtain a customer service decoding result, and the decoding of the visitor voice sequence to obtain a visitor decoding result, include:
[0010] The customer service voice sequence is initially decoded to obtain each customer service candidate decoding sequence, and the visitor voice sequence is initially decoded to obtain each visitor candidate decoding sequence.
[0011] Each of the aforementioned customer service candidate decoding sequences is initially scored to obtain a customer service decoding score, and each of the aforementioned visitor candidate decoding sequences is initially scored to obtain a visitor decoding score.
[0012] Determine the customer service decoding sequence with the highest customer service decoding score, and obtain the customer service decoding result corresponding to the customer service candidate sequence;
[0013] Determine the visitor decoding sequence with the highest visitor decoding score, and obtain the visitor decoding result corresponding to the visitor decoding sequence.
[0014] In one possible implementation of this application, the step of rescoring the customer service voice sequence based on the customer service decoding result and rescoring the visitor voice sequence based on the visitor decoding result includes:
[0015] The customer service candidate decoding sequence is further filtered to obtain a customer service filter word grid, and the visitor candidate decoding sequence is further filtered by path to obtain a visitor filter word grid.
[0016] The customer service candidate decoding paths in the customer service filter word grid and the customer service decoding results are input into a preset rescoring model to calculate the secondary customer service score of the customer service candidate decoding paths;
[0017] The visitor candidate decoding paths in the visitor filter word grid and the visitor decoding results are input into the rescoring model to calculate the visitor secondary score of the visitor candidate decoding paths.
[0018] In one possible implementation of this application, the step of performing secondary decoding on the re-scored customer service voice sequence and secondary decoding on the re-scored visitor voice sequence to obtain the speech recognition result of the voice conversation information includes:
[0019] Obtain the customer service starting voice sequence of the customer service candidate decoding path, and obtain the reference customer service voice sequence associated with the customer service starting voice sequence in the customer service decoding result;
[0020] The customer service candidate decoding path is decoded to obtain the customer service decoding candidate result;
[0021] Obtain the reference decoding result corresponding to the preset sub-sequence of the reference customer service voice sequence, and associate and concatenate the reference decoding result and the customer service decoding candidate result to obtain the concatenation candidate result;
[0022] A secondary scoring calculation is performed on the splicing candidate results to obtain the secondary customer service score of the customer service candidate decoding path.
[0023] In one possible implementation of this application, filtering the synthesized audio in the voice conversation information to be identified includes:
[0024] The initial voice conversation information is processed into frames to obtain customer service voice frames;
[0025] Voice endpoint detection is performed on the customer service voice frames to determine the silence segmentation point of the voice conversation information;
[0026] The voice conversation information is segmented into silence based on the silence segmentation point to determine the silent voice sequence in the voice conversation information;
[0027] Filter the silent speech sequences in the voice conversation information to obtain the filtered voice conversation information.
[0028] In one possible implementation of this application, filtering the synthesized audio in the voice conversation information to be identified includes:
[0029] Obtain the voiceprint feature vector of the speech conversation information to be identified;
[0030] The similarity between the voiceprint feature vector and the preset human voice voiceprint vector is performed to obtain the voiceprint similarity of the voiceprint feature vector;
[0031] The voiceprint similarity is compared with a preset human voice matching threshold. If the voiceprint similarity is less than the human voice matching threshold, the voice conversation information is determined to be synthetic audio, and the synthetic audio is filtered out.
[0032] In one possible implementation of this application, the step of clustering the filtered voice conversation information to obtain customer service voice sequences and visitor voice sequences includes:
[0033] Read customer service interaction information and preset the number of speakers in the voice conversation information based on the customer service interaction information;
[0034] Obtain each voiceprint feature vector from the filtered voice conversation information, and parse the voiceprint feature vector to obtain the speaker information of the voiceprint feature vector;
[0035] Based on the number of speakers and the speaker information, the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0036] On the other hand, this application provides a customer service voice recognition device, the customer service voice recognition device comprising:
[0037] The voice acquisition module is configured to acquire voice conversation information to be recognized;
[0038] The speaker separation module is configured to filter the synthesized audio in the voice conversation information, and cluster the filtered voice conversation information to obtain customer service voice sequences and visitor voice sequences.
[0039] The voice decoding module is configured to decode the customer service voice sequence to obtain the customer service decoding result; and to decode the visitor voice sequence to obtain the visitor decoding result.
[0040] The secondary decoding module is configured to re-score the customer service voice sequence based on the customer service decoding result, and re-score the visitor voice sequence based on the visitor decoding result, to obtain the speech recognition result of the voice conversation information.
[0041] On the other hand, this application also provides a customer service voice recognition device, the customer service voice recognition device comprising:
[0042] One or more processors;
[0043] Memory; and
[0044] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the customer service voice recognition method.
[0045] On the other hand, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps in the customer service voice recognition method.
[0046] This application acquires the voice conversation information to be recognized; detects and filters the synthesized audio in the voice conversation information to avoid interference with the speech recognition results; clusters the filtered voice conversation information to separate the speakers, obtaining customer service voice sequences and visitor voice sequences; after obtaining the customer service voice sequences and visitor voice sequences, the customer service voice sequences are initially decoded to obtain customer service decoding results; the visitor voice sequences are initially decoded to obtain visitor decoding results; and the customer service voice sequences are re-scored based on the customer service decoding results, and the visitor voice sequences are re-scored based on the visitor decoding results. The re-scored customer service voice sequences are then decoded a second time, and the re-scored visitor voice sequences are also decoded a second time to obtain the speech recognition results of the voice conversation information. This achieves secondary decoding of the separated customer service and visitor voices, improving the speech recognition accuracy in customer service voice scenarios. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0048] Figure 1 This is a schematic diagram illustrating a scenario of the customer service voice recognition method in an embodiment of this application.
[0049] Figure 2 This is a flowchart illustrating one embodiment of the customer service voice recognition method in this application.
[0050] Figure 3 A flowchart illustrating an embodiment of the customer service voice recognition method provided in this application, which involves rescoring a voice sequence.
[0051] Figure 4 This is a flowchart illustrating one embodiment of the customer service voice recognition method for calculating secondary customer service scores for candidate decoding paths in this application.
[0052] Figure 5 A schematic diagram of one embodiment of the customer service voice recognition device provided in this application;
[0053] Figure 6 This is a schematic diagram of an embodiment of the customer service voice recognition device provided in this application. Detailed Implementation
[0054] The technical solutions of the embodiments of this application 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.
[0055] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0056] In this application, the term "exemplary" is used to mean "serving as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0057] Currently, with the continuous development of speech recognition technology, many industries have developed intelligent voice customer service systems based on this technology. These systems convert speech to text using speech recognition and then analyze the text using natural language processing algorithms to extract key information from customer service conversations. However, existing speech recognition systems use endpoint detection and speaker separation to segment speech into sequences before recognizing them, resulting in the loss of contextual information and inaccurate speech recognition. Secondly, customer service voices are often synthesized, and existing intelligent voice customer service systems do not consider synthesized voices during operation, leading to reduced accuracy in speaker separation and decoding, making it difficult to meet the needs of current customer service scenarios.
[0058] Based on this, this application proposes a customer service voice recognition method, apparatus, device, and computer-readable storage medium to solve the technical problem of poor accuracy in voice recognition and speaker separation in intelligent voice customer service systems.
[0059] The customer service voice recognition method in this embodiment of the invention is applied to a customer service voice recognition device. The customer service voice recognition device is set in a customer service voice recognition equipment. The customer service voice recognition equipment is provided with one or more processors, a memory, and one or more applications. The one or more applications are stored in the memory and configured to be executed by the processor to implement the customer service voice recognition method. The customer service voice recognition equipment can be a smart terminal, such as a mobile phone, tablet computer, smart TV, network device, and smart computer. Optionally, the customer service voice recognition equipment can also be a server or a service cluster composed of multiple servers.
[0060] like Figure 1 As shown, Figure 1 This is a schematic diagram of a customer service voice recognition method according to an embodiment of this application. The customer service voice recognition scenario in this embodiment includes a customer service voice recognition device 100 (the customer service voice recognition device 100 integrates a customer service voice recognition apparatus). The customer service voice recognition device 100 is equipped with a computer-readable storage medium corresponding to the customer service voice recognition method to execute the steps of the customer service voice recognition method.
[0061] Understandable, Figure 1 The customer service voice recognition device in the customer service voice recognition method scenario shown, or the device included in the customer service voice recognition device, does not constitute a limitation on the embodiments of the present invention. That is, the number or type of customer service voice recognition device included in the customer service voice recognition method scenario, or the number or type of device included in each device, does not affect the overall implementation of the technical solution in the embodiments of the present invention, and can all be considered as equivalent substitutions or derivatives of the technical solutions claimed in the embodiments of the present invention.
[0062] In this embodiment of the invention, the customer service voice recognition device 100 is mainly used for:
[0063] Synthetic audio is filtered from the voice conversation information to be identified, and the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0064] The customer service voice sequence is initially decoded to obtain the customer service decoding result, and the visitor voice sequence is initially decoded to obtain the visitor decoding result.
[0065] The customer service voice sequence is re-scored based on the customer service decoding result, and the visitor voice sequence is re-scored based on the visitor decoding result;
[0066] The customer service voice sequence after rescoring is decoded a second time, and the visitor voice sequence after rescoring is decoded a second time to obtain the voice recognition result of the voice conversation information.
[0067] The customer service voice recognition device 100 in this embodiment of the invention can be an independent customer service voice recognition device, such as a mobile phone, tablet computer, smart TV, network device, server and smart computer, or a customer service voice recognition network or customer service voice recognition cluster composed of multiple customer service voice recognition devices.
[0068] This application provides a customer service voice recognition method, apparatus, device, and computer-readable storage medium, which will be described in detail below.
[0069] It will be understood by those skilled in the art that Figure 1 The application environment shown is only one application scenario related to the solution of this application and does not constitute a limitation on the application scenario of this application. Other application environments may include more than one application scenario. Figure 1 The number of customer service voice recognition devices shown, or the customer service voice recognition network connectivity, for example... Figure 1 Only one customer service voice recognition device is shown in the document. It is understood that the scenario of this customer service voice recognition method may also include one or more customer service voice recognition devices, which are not specifically limited here. The customer service voice recognition device may also include a memory for storing voice conversation information and other data.
[0070] It should be noted that, Figure 1 The schematic diagram of the customer service voice recognition method shown is merely an example. The scenarios of the customer service voice recognition method described in this embodiment are for the purpose of more clearly illustrating the technical solutions of this embodiment and do not constitute a limitation on the technical solutions provided by this embodiment.
[0071] Based on the scenarios described above for customer service voice recognition methods, various embodiments of the customer service voice recognition method disclosed in this invention are proposed.
[0072] like Figure 2 As shown, Figure 2 This is a flowchart illustrating one embodiment of the customer service voice recognition method in this application. The customer service voice recognition method includes the following steps 201 to 203:
[0073] 201. Filter the synthesized audio in the speech conversation information to be identified, and cluster the filtered speech conversation information to obtain the customer service speech sequence and the visitor speech sequence;
[0074] The customer service voice recognition method in this embodiment is applied to a customer service voice recognition device. The type and number of customer service voice recognition devices are not specifically limited. That is, the customer service voice recognition device can be one or more smart terminals or servers. In a specific embodiment, the customer service voice recognition device is a smart computer.
[0075] Specifically, the customer service voice recognition device is configured as an intelligent voice customer service system, generating corresponding customer service conversations for visitors and customer service representatives, enabling them to communicate and interact within these conversations. During operation, the customer service voice recognition device receives and responds to customer service conversation voice recognition requests, performing voice recognition on the customer service voice to be recognized. These requests can be user-initiated, for example, by a customer service representative or manager clicking the customer service voice recognition button on the device. The customer service conversation voice recognition request is the operational instruction that drives the device to collect voice conversation information and perform voice recognition on the voice conversation information to be recognized. Optionally, the customer service conversation voice recognition request can also be automatically triggered by the device. For example, the device may have a pre-set voice recognition process; after a customer service conversation is created, it automatically triggers the request, collects the voice conversation information, and performs voice recognition on that information.
[0076] Specifically, after receiving a customer service conversation voice recognition request, the customer service voice recognition device monitors the customer service conversation and collects the voice conversation information within it. This voice conversation information is generated from the voice communication interaction between the customer service representative and the visitor, carrying voice signals from both the customer service representative and the visitor. After acquiring the initial voice conversation information, to improve decoding speed and accuracy, the customer service voice recognition device performs voice endpoint detection on the voice conversation information, thereby obtaining several silent voice sequences within the initial voice conversation information. These silent voice sequences are then filtered out to obtain the voice conversation information to be recognized.
[0077] Optionally, when customer service representatives connect to a customer service session, they may use synthesized voices to communicate with visitors. When the customer service voice recognition device collects voice session information, this information also includes the synthesized voice signal, which can interfere with the customer service voice recognition. The customer service voice recognition device performs voice endpoint detection on the voice session information, filters out silent speech sequences, and then performs voiceprint verification on the filtered voice session information to filter out the synthesized audio.
[0078] Specifically, the customer service voice recognition device is pre-configured with a voiceprint verification model. In one specific embodiment, this voiceprint verification model includes an i-vector model and a PLDA model. The customer service voice recognition device pre-trains this voiceprint verification model using a synthesized speech dataset and a human voice dataset to obtain a human voice voiceprint vector. This human voice voiceprint vector is a vector representing the speaker's characteristics. The formula for calculating the voiceprint vector is s = m + Tw, where s is the voiceprint feature, m is the hypervector, w is the identity vector, and T is the total variable space. After obtaining the human voice voiceprint vector, the PLDA model is used to refine the human voice voiceprint vector.
[0079] The customer service voice recognition device performs appointment endpoint detection on the voice conversation information, filters out silent speech sequences, and then inputs the filtered voice conversation information into the voiceprint verification model to calculate the voiceprint feature vectors of the voice conversation information. The device then performs similarity detection between these voiceprint feature vectors and the preset human voiceprint vectors in the voiceprint verification model to obtain the voiceprint similarity of the voiceprint feature vectors.
[0080] After obtaining the voiceprint similarity of the voiceprint feature vector, the customer service voice recognition device compares the voiceprint similarity with a preset human voice matching threshold to determine whether the part of the speech conversation information corresponding to the voiceprint feature vector is synthetic audio. The human voice matching threshold is a similarity threshold obtained during training in the voiceprint verification model for detecting whether a speech sequence is a synthetic audio signal.
[0081] Optionally, if the voiceprint similarity is greater than the human voice matching threshold, the part of the voice conversation information corresponding to the voiceprint similarity is determined to be a human voice speech sequence, and the customer service voice recognition device is controlled to further separate the human voice speech sequence.
[0082] Optionally, if the voiceprint similarity is less than the human voice matching threshold, then the effective speech sequence corresponding to the voiceprint similarity is determined to be a synthesized audio, and the synthesized audio is filtered.
[0083] After verifying the voiceprints of all voice conversations and filtering out silent speech sequences and synthesized audio, the customer service voice recognition device further separates the speakers in the filtered voice conversations. This distinguishes between customer service voice sequences and visitor voice sequences within the voice conversation information.
[0084] When creating and monitoring a customer service session, the customer service voice recognition device also acquires the customer service interaction information of that session and presets the number of speakers in the session based on this information. This customer service interaction information includes data representing the number of customer service representatives, customer service work information, and the number of visitors in the session.
[0085] Specifically, after acquiring customer service interaction information, the customer service voice recognition device reads the number of customer service participants and the number of visitor participants in the interaction information to determine the number of speakers in the customer service session, and sets this number of speakers as the number of clusters for speaker separation. Optionally, in a specific embodiment, if the number of customer service participants and the number of visitor participants in a customer service session's interaction information are 1, then the customer service voice recognition device determines the number of speakers in the customer service session to be 2, that is, the customer service voice recognition device sets the number of clusters for the customer service session information to be 2.
[0086] After determining the number of speakers, the customer service voice recognition device obtains the voiceprint feature vectors of each voiceprint in the voiceprint verification model, and determines the speaker information based on the identity vector in the voiceprint feature vector.
[0087] Specifically, after determining the number of speakers and the identity vector of the voiceprint feature vector, the customer service voice recognition device clusters the voiceprint feature vector according to the speaker information corresponding to the identity vector. It traverses the voiceprint feature vector and clusters the voiceprint feature vectors with the same identity vector, thereby clustering them into a number of customer service voice sequences and visitor voice sequences equal to the number of speakers, thus achieving speaker separation of voice conversation information.
[0088] 202. Perform initial decoding on the customer service voice sequence to obtain the customer service decoding result; perform initial decoding on the visitor voice sequence to obtain the visitor decoding result.
[0089] After the customer service voice recognition device separates the speakers from the voice conversation information to obtain the customer service voice sequence and the visitor voice sequence, it performs initial decoding on the customer service voice sequence and the visitor voice sequence to obtain the customer service decoding result of the customer service voice sequence and the visitor decoding result of the visitor voice sequence.
[0090] Specifically, the customer service voice recognition device is pre-set with a rescoring model, which includes a first decoding module and a rescoring module. Optionally, in one specific embodiment, the first decoding module is an HCLG.fst decoder. After acquiring the customer service voice sequence and the visitor voice sequence, the customer service voice device inputs the customer service voice sequence and the visitor voice sequence into the first decoding module for decoding, respectively, to obtain the customer service candidate decoding sequences corresponding to the customer service voice sequence and the visitor candidate decoding sequences corresponding to the visitor voice sequence.
[0091] After the customer service voice recognition device inputs the customer service voice sequence and the visitor voice sequence into the first decoding module, it calls the first decoding module to calculate the customer service decoding score for each candidate customer service decoding sequence and the visitor decoding score for each candidate visitor decoding sequence. Specifically, the customer service decoding device calls the first decoding module to obtain the graph cost and acoustic cost for each candidate customer service decoding sequence, and obtains the preset graph cost weight and acoustic cost weight of the first decoding module. Based on the obtained graph cost, acoustic cost, graph cost weight, and acoustic cost weight, it calculates the customer service decoding score for the candidate customer service decoding path and the visitor decoding score for the candidate visitor decoding path. Specifically, the calculation formula for the customer service decoding score and the visitor decoding score is: Cost = lmwt * graph_cost + acwt * acoustic_cost. Where lmwt is the graph cost weight, graph_cost is the graph cost, acwt is the acoustic cost weight, and acoustic_cost is the acoustic cost.
[0092] After obtaining the customer service decoding score and the visitor decoding score, the customer service voice recognition device compares the customer service decoding score and the visitor decoding score, obtains the customer service decoding sequence with the highest customer service decoding score, and obtains the customer service decoding result corresponding to the customer service decoding sequence; and determines the visitor decoding sequence with the highest visitor decoding score and the visitor decoding result corresponding to the visitor decoding sequence.
[0093] After the customer service voice recognition device performs the initial decoding of the customer service voice sequence and the visitor voice sequence, it uses the visitor decoding result and the customer service decoding result as reference decoding results, and then re-scores and performs secondary decoding on the customer service voice sequence and the visitor voice sequence based on the reference decoding results.
[0094] 203. Re-scoring the customer service voice sequence based on the customer service decoding result, and re-scoring the visitor voice sequence based on the visitor decoding result;
[0095] After initially decoding the customer service voice sequence and the visitor voice sequence, the customer service voice recognition device uses the visitor decoding result and the customer service decoding result as reference decoding results, and re-scores the customer service voice sequence and the visitor voice sequence based on the reference decoding results.
[0096] Specifically, the customer service voice recognition device inputs the customer service decoding result and the customer service voice sequence into the rescoring module of the rescoring model to obtain the secondary customer service score corresponding to the customer service voice sequence. The visitor voice recognition device inputs the visitor decoding result and the visitor voice sequence into the rescoring module of the rescoring model to obtain the secondary visitor score corresponding to the visitor voice sequence.
[0097] 204. Perform secondary decoding on the re-scored customer service voice sequence and the re-scored visitor voice sequence to obtain the voice recognition result of the voice conversation information.
[0098] After acquiring the re-scored customer service voice sequence and the re-scored visitor voice sequence, the customer service voice recognition device performs secondary decoding on the re-scored customer service voice sequence to obtain the target customer service decoding result; and performs secondary decoding on the re-scored visitor voice sequence to obtain the target visitor decoding result.
[0099] After obtaining the decoding results of the target customer service representative and the target visitor, the customer service voice recognition device summarizes the voice recognition results of the voice conversation information.
[0100] This application acquires the voice conversation information to be recognized; detects and filters the synthesized audio in the voice conversation information to avoid interference from synthesized audio on the speech recognition results; clusters the filtered voice conversation information to separate the speakers, obtaining customer service voice sequences and visitor voice sequences; after obtaining the customer service voice sequences and visitor voice sequences, decodes the customer service voice sequences to obtain customer service decoding results; decodes the visitor voice sequences to obtain visitor decoding results; and performs secondary decoding on the customer service voice sequences based on the customer service decoding results, re-scoring the customer service voice sequences, and re-scoring the visitor voice sequences based on the visitor decoding results to obtain the speech recognition results of the voice conversation information. This achieves effective recognition and filtering of synthesized audio in customer service voice scenarios containing synthesized audio, thereby improving the accuracy of speaker separation, and further improves the speech recognition accuracy in customer service voice scenarios by performing secondary decoding on the separated customer service and visitor voices.
[0101] like Figure 3 As shown, Figure 3 This is a flowchart illustrating an embodiment of the customer service voice recognition method provided in this application, which involves rescoring the voice sequence.
[0102] Based on the above embodiments, the speech recognition results of the voice conversation information are obtained by re-scoring the customer service voice sequence and the visitor voice sequence in this application embodiment. Specifically, it includes steps 301 to 303:
[0103] 301. Further filter the customer service candidate decoding sequence to obtain a customer service filter word grid, and further filter the visitor candidate decoding sequence by path to obtain a visitor filter word grid;
[0104] 302. Input the customer service candidate decoding path in the customer service filter word grid and the customer service decoding result into the preset rescoring model, and calculate the customer service secondary score of the customer service candidate decoding path;
[0105] 303. Input the visitor candidate decoding path in the visitor filter word grid and the visitor decoding result into the rescoring model, and calculate the visitor secondary score of the visitor candidate decoding path.
[0106] In this embodiment, the customer service speech recognition device further filters the customer service candidate decoding sequences to obtain a customer service filter word grid, and further filters the visitor candidate decoding sequences to obtain a visitor filter word grid. Specifically, the customer service speech recognition device obtains the N-BEST results of each customer service candidate decoding sequence and generates the customer service filter word grid; the customer service speech recognition device obtains the N-BEST results of each visitor candidate decoding sequence and generates the visitor filter word grid.
[0107] After acquiring the customer service filter word grid, the customer service speech recognition device obtains a reference decoding result for the customer service decoding result. This reference result and the customer service filter word grid are then input into the rescoring module of the rescoring model to rescorize the customer service filter word grid, thus obtaining the final customer service speech decoding result. This rescoring module is a recurrent neural network model.
[0108] Specifically, when the customer service voice recognition device obtains the customer service decoding result, it also obtains the customer service decoding score corresponding to the customer service decoding result, and gives a secondary customer service score to the customer service candidate decoding path based on the customer service decoding score.
[0109] Specifically, the customer service voice recognition device inputs the customer service candidate decoding paths and customer service decoding scores of each customer service candidate decoding path into the rescoring module to calculate the secondary customer service score for each customer service candidate decoding path.
[0110] Specifically, the formula for calculating customer service secondary ratings or visitor secondary ratings is as follows:
[0111] New_cost = lmwt * (graph_cost - g + g1) + acwt * acoustic_cost, where g is the customer service decoding score, g1 is the recurrent neural network model score of the candidate decoding path, and New_cost is the customer service secondary score or visitor secondary score.
[0112] Specifically, when the visitor voice recognition device obtains the visitor decoding result, it also obtains the visitor decoding score corresponding to the visitor decoding result, and scores the visitor secondary score of the visitor candidate decoding path based on the visitor decoding score.
[0113] Specifically, the visitor speech recognition device inputs each visitor candidate decoding path and visitor decoding score into the rescoring module to calculate the visitor secondary score for each visitor candidate decoding path.
[0114] After calculating the visitor secondary score for all candidate visitor decoding paths and the customer service secondary score for all candidate customer service paths, the visitor candidate decoding path with the highest visitor secondary score is determined as the target visitor decoding path, and the customer service candidate decoding path with the highest customer service secondary score is determined as the target customer service decoding path. The target customer service decoding path and the target visitor decoding path are then decoded to obtain the speech recognition result of the voice conversation information.
[0115] Optionally, when rescoring the customer service voice sequence and the visitor voice sequence, the customer service voice recognition device further uses the customer service decoding result and the visitor decoding result as reference decoding results to rescorate together with the customer service voice sequence and the visitor voice sequence, thereby improving the accuracy of the recognition results.
[0116] In this embodiment, the customer service speech recognition device further filters the customer service candidate decoding sequence to obtain a customer service filter word grid, and further filters the visitor candidate decoding sequence to obtain a visitor filter word grid. The customer service candidate decoding paths in the customer service filter word grid and the customer service decoding results are input into a preset rescoring model to calculate the secondary customer service score of the customer service candidate decoding paths. The visitor candidate decoding paths in the visitor filter word grid and the visitor decoding results are input into the rescoring model to calculate the secondary visitor score of the visitor candidate decoding paths. Based on the secondary customer service score and the secondary visitor score, a target customer service decoding path and a target visitor decoding path are determined. The target customer service decoding path and the target visitor decoding path are decoded to obtain the speech recognition result of the voice conversation information. This improves the accuracy of customer service speech recognition.
[0117] like Figure 4 As shown, Figure 4 This is a flowchart illustrating one embodiment of the customer service voice recognition method for calculating secondary customer service scores for candidate decoding paths in this application.
[0118] Based on the above embodiments, this embodiment obtains a reference decoding result corresponding to the customer service decoding result, and calculates a secondary customer service score for the customer service candidate decoding path based on the reference decoding result. Specifically, it includes steps 401 to 404:
[0119] 401. Obtain the customer service starting voice sequence of the customer service candidate decoding path, and obtain the reference customer service voice sequence associated with the customer service starting voice sequence in the customer service decoding result;
[0120] 402. Decode the customer service candidate decoding path to obtain the customer service decoding candidate result;
[0121] 403. Obtain the reference decoding result corresponding to the preset sub-sequence of the reference customer service voice sequence, and associate and concatenate the reference decoding result and the customer service decoding candidate result to obtain the concatenation candidate result;
[0122] 404. Perform a secondary scoring calculation on the splicing candidate results to obtain the secondary customer service score of the customer service candidate decoding path.
[0123] In this embodiment, when the customer service voice recognition device re-scores the customer service voice sequence and the visitor voice sequence, it also obtains the customer service starting voice sequence for each customer service candidate decoding path in the customer service filter word grid. This starting voice sequence consists of the first few voice sequences of each customer service candidate decoding path. After obtaining the starting voice sequence, the customer service voice recognition device also obtains a reference customer service voice sequence associated with it in the customer service decoding result. This reference customer service voice sequence is the customer service decoding sequence located at the same position as the starting voice sequence.
[0124] Specifically, the customer service speech recognition device inputs the candidate decoding path into the rescoring module of the rescoring model for secondary decoding to obtain the candidate decoding result. The device further acquires reference decoding results corresponding to preset sub-sequences of the reference customer service speech sequence and appends these reference decoding results to the beginning of the reference decoding result as a contextual reference for the candidate decoding result, generating a concatenated candidate result. The sub-sequence consists of several speech sequences preceding the starting speech sequence of the candidate decoding path. The length of this sub-sequence can be customized according to actual decoding speed requirements.
[0125] The customer service voice recognition device performs a secondary scoring calculation on the concatenated candidate result to obtain a secondary score for the customer service candidate decoding path. This improves the recognition accuracy of shorter customer service candidate decoding paths.
[0126] In this embodiment, the customer service speech recognition device obtains the initial customer service speech sequence of the customer service candidate decoding path, and obtains a reference customer service speech sequence associated with the initial customer service speech sequence in the customer service decoding result; decodes the customer service candidate decoding path to obtain customer service decoding candidate results; obtains a reference decoding result corresponding to a preset sub-sequence of the reference customer service speech sequence, and concatenates the reference decoding result and the customer service decoding candidate results as a contextual reference for the customer service decoding candidate results to obtain a concatenated candidate result; performs a secondary scoring calculation on the concatenated candidate result to obtain a secondary evaluation score for the customer service candidate decoding path. This improves the decoding accuracy of customer service speech recognition.
[0127] To better implement the customer service voice recognition method in the embodiments of this application, based on the customer service voice recognition method, the embodiments of this application also provide a customer service voice recognition device, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of one embodiment of the customer service voice recognition device provided in this application. The customer service voice recognition device 500 includes:
[0128] The synthesis filtering module 501 is configured to filter the synthesized audio in the voice conversation information to be recognized, and to cluster the filtered voice conversation information to obtain customer service voice sequence and visitor voice sequence.
[0129] The initial decoding module 502 is configured to perform initial decoding on the customer service voice sequence to obtain the customer service decoding result, and to perform initial decoding on the visitor voice sequence to obtain the visitor decoding result.
[0130] The secondary scoring module 503 is configured to re-score the customer service voice sequence based on the customer service decoding result, and to re-score the visitor voice sequence based on the visitor decoding result;
[0131] The secondary decoding module 504 is configured to perform secondary decoding on the re-scored customer service voice sequence and on the re-scored visitor voice sequence to obtain the voice recognition result of the voice conversation information.
[0132] In some embodiments of this application, the customer service voice recognition device performs initial decoding on the customer service voice sequence to obtain a customer service decoding result; and decodes the visitor voice sequence to obtain a visitor decoding result, including:
[0133] The customer service voice sequence is initially decoded to obtain each customer service candidate decoding sequence, and the visitor voice sequence is initially decoded to obtain each visitor candidate decoding sequence.
[0134] Each of the aforementioned customer service candidate decoding sequences is initially scored to obtain a customer service decoding score, and each of the aforementioned visitor candidate decoding sequences is initially scored to obtain a visitor decoding score.
[0135] Determine the customer service decoding sequence with the highest customer service decoding score, and obtain the customer service decoding result corresponding to the customer service candidate sequence;
[0136] Determine the visitor decoding sequence with the highest visitor decoding score, and obtain the visitor decoding result corresponding to the visitor decoding sequence.
[0137] In some embodiments of this application, the customer service voice recognition device re-scores the customer service voice sequence based on the customer service decoding result, and re-scores the visitor voice sequence based on the visitor decoding result, including:
[0138] The customer service candidate decoding sequence is further filtered to obtain a customer service filter word grid, and the visitor candidate decoding sequence is further filtered by path to obtain a visitor filter word grid.
[0139] The customer service candidate decoding paths in the customer service filter word grid and the customer service decoding results are input into a preset rescoring model to calculate the secondary customer service score of the customer service candidate decoding paths;
[0140] The visitor candidate decoding paths in the visitor filter word grid and the visitor decoding results are input into the rescoring model to calculate the visitor secondary score of the visitor candidate decoding paths.
[0141] In some embodiments of this application, a customer service voice recognition device
[0142] The customer service voice sequence after rescoring is decoded a second time, and the visitor voice sequence after rescoring is decoded a second time to obtain the speech recognition result of the voice conversation information, including:
[0143] Obtain the customer service starting voice sequence of the customer service candidate decoding path, and obtain the reference customer service voice sequence associated with the customer service starting voice sequence in the customer service decoding result;
[0144] The customer service candidate decoding path is decoded to obtain the customer service decoding candidate result;
[0145] Obtain the reference decoding result corresponding to the preset sub-sequence of the reference customer service voice sequence, and associate and concatenate the reference decoding result and the customer service decoding candidate result to obtain the concatenation candidate result;
[0146] A secondary scoring calculation is performed on the splicing candidate results to obtain the secondary customer service score of the customer service candidate decoding path.
[0147] In some embodiments of this application, the customer service voice recognition device filters synthetic audio from the voice conversation information to be recognized.
[0148] In some embodiments of this application, the customer service voice recognition device filters synthetic audio from the voice conversation information to be recognized, including:
[0149] Obtain the voiceprint feature vector of the speech conversation information to be identified;
[0150] The similarity between the voiceprint feature vector and the preset human voice voiceprint vector is performed to obtain the voiceprint similarity of the voiceprint feature vector;
[0151] The voiceprint similarity is compared with a preset human voice matching threshold. If the voiceprint similarity is less than the human voice matching threshold, the voice conversation information is determined to be synthetic audio, and the synthetic audio is filtered out.
[0152] In some embodiments of this application, the customer service voice recognition device clusters the filtered voice conversation information to obtain customer service voice sequences and visitor voice sequences, including:
[0153] Read customer service interaction information and preset the number of speakers in the voice conversation information based on the customer service interaction information;
[0154] Obtain each voiceprint feature vector from the filtered voice conversation information, and parse the voiceprint feature vector to obtain the speaker information of the voiceprint feature vector;
[0155] Based on the number of speakers and the speaker information, the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0156] In this embodiment, the customer service voice recognition device acquires the voice conversation information to be recognized; detects and filters the synthesized audio in the voice conversation information to avoid interference with the voice recognition result; clusters the filtered voice conversation information to separate the speaker, obtaining a customer service voice sequence and a visitor voice sequence; after obtaining the customer service voice sequence and the visitor voice sequence, decodes the customer service voice sequence to obtain a customer service decoding result; decodes the visitor voice sequence to obtain a visitor decoding result; and performs secondary decoding on the customer service voice sequence based on the customer service decoding result, re-scoring the customer service voice sequence, and re-scoring the visitor voice sequence based on the visitor decoding result to obtain the voice recognition result of the voice conversation information. This effectively identifies and filters synthesized audio in customer service voice scenarios containing synthesized audio, thereby improving the accuracy of speaker separation, and further improves the voice recognition accuracy in customer service voice scenarios by performing secondary decoding on the separated customer service and visitor voices.
[0157] This invention also provides a customer service voice recognition device, such as... Figure 6 As shown, Figure 6 This is a schematic diagram of an embodiment of the customer service voice recognition device provided in this application.
[0158] The customer service voice recognition device integrates any of the customer service voice recognition devices provided in the embodiments of the present invention, and the customer service voice recognition device includes:
[0159] One or more processors;
[0160] Memory; and
[0161] One or more applications, wherein the one or more applications are stored in the memory and configured by the processor to perform the steps of the customer service voice recognition method described in any of the embodiments of the above customer service voice recognition method.
[0162] Specifically, a customer service voice recognition device may include components such as a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, a power supply 603, and an input unit 604. Those skilled in the art will understand that... Figure 6 The customer service voice recognition device structure shown does not constitute a limitation on the device and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. Wherein:
[0163] The processor 601 is the control center of the customer service voice recognition device. It connects various parts of the device via interfaces and lines, and executes software programs and / or modules stored in the memory 602, as well as calling data stored in the memory 602, to perform various functions and process data, thereby providing overall monitoring of the device. Optionally, the processor 601 may include one or more processing cores; preferably, it may integrate an application processor and a modem processor, where the application processor primarily handles the operating system, user interface, and applications, while the modem processor primarily handles wireless communication. It is understood that the modem processor may not be integrated into the processor 601.
[0164] The memory 602 can be used to store software programs and modules. The processor 601 executes various functional applications and data processing by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the customer service voice recognition device, etc. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide the processor 601 with access to the memory 602.
[0165] The customer service voice recognition device also includes a power supply 603 that supplies power to the various components. Preferably, the power supply 603 can be logically connected to the processor 601 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 603 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.
[0166] The customer service voice recognition device may also include an input unit 604, which can be used to receive input numeric or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
[0167] Although not shown, the customer service voice recognition device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 601 in the customer service voice recognition device loads the executable files corresponding to the processes of one or more applications into the memory 602 according to the following instructions, and the processor 601 runs the applications stored in the memory 602 to realize various functions, as follows:
[0168] Synthetic audio is filtered from the voice conversation information to be identified, and the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0169] The customer service voice sequence is initially decoded to obtain the customer service decoding result, and the visitor voice sequence is initially decoded to obtain the visitor decoding result.
[0170] The customer service voice sequence is re-scored based on the customer service decoding result, and the visitor voice sequence is re-scored based on the visitor decoding result;
[0171] The customer service voice sequence after rescoring is decoded a second time, and the visitor voice sequence after rescoring is decoded a second time to obtain the voice recognition result of the voice conversation information.
[0172] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0173] Therefore, embodiments of the present invention provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), a disk, or an optical disk, etc. A computer program is stored thereon, which is loaded by a processor to execute the steps in any of the customer service voice recognition methods provided in the embodiments of the present invention. For example, the computer program loaded by the processor can execute the following steps:
[0174] Synthetic audio is filtered from the voice conversation information to be identified, and the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
[0175] The customer service voice sequence is initially decoded to obtain the customer service decoding result, and the visitor voice sequence is initially decoded to obtain the visitor decoding result.
[0176] The customer service voice sequence is re-scored based on the customer service decoding result, and the visitor voice sequence is re-scored based on the visitor decoding result;
[0177] The customer service voice sequence after rescoring is decoded a second time, and the visitor voice sequence after rescoring is decoded a second time to obtain the voice recognition result of the voice conversation information.
[0178] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the detailed descriptions of other embodiments above, which will not be repeated here.
[0179] In practice, each of the above units or structures can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units or structures, please refer to the previous method embodiments, which will not be repeated here.
[0180] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0181] The above provides a detailed description of a customer service voice recognition method provided by the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A customer service voice recognition method, characterized in that, The customer service voice recognition method includes: Synthetic audio is filtered from the voice conversation information to be identified, and the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences. The customer service voice sequence is initially decoded to obtain the customer service decoding result, and the visitor voice sequence is initially decoded to obtain the visitor decoding result. The customer service voice sequence is re-scored based on the customer service decoding result, and the visitor voice sequence is re-scored based on the visitor decoding result; Obtain the customer service starting voice sequence of the customer service candidate decoding path; obtain the reference customer service voice sequence associated with the customer service starting voice sequence in the customer service decoding result; decode the customer service candidate decoding path to obtain customer service decoding candidate results; obtain the reference decoding result corresponding to the preset sub-sequence of the reference customer service voice sequence; associate and concatenate the reference decoding result and the customer service decoding candidate results to obtain the concatenated candidate result; perform a secondary scoring calculation on the concatenated candidate result to obtain the customer service secondary score of the customer service candidate decoding path. The customer service candidate decoding path is obtained by further filtering the customer service candidate decoding sequence; the customer service candidate decoding sequence is obtained by initially decoding the customer service voice sequence.
2. The customer service voice recognition method as described in claim 1, characterized in that, The customer service voice sequence is initially decoded to obtain the customer service decoding result. Decoding the visitor's voice sequence yields the visitor decoding result, including: The customer service voice sequence is initially decoded to obtain each customer service candidate decoding sequence, and the visitor voice sequence is initially decoded to obtain each visitor candidate decoding sequence. Each of the aforementioned customer service candidate decoding sequences is initially scored to obtain a customer service decoding score, and each of the aforementioned visitor candidate decoding sequences is initially scored to obtain a visitor decoding score. Determine the customer service decoding sequence with the highest customer service decoding score, and obtain the customer service decoding result corresponding to the customer service decoding sequence; Determine the visitor decoding sequence with the highest visitor decoding score, and obtain the visitor decoding result corresponding to the visitor decoding sequence.
3. The customer service voice recognition method as described in claim 2, characterized in that, The step of re-scoring the customer service voice sequence based on the customer service decoding result, and re-scoring the visitor voice sequence based on the visitor decoding result, includes: The customer service candidate decoding sequence is further filtered to obtain a customer service filter word grid, and the visitor candidate decoding sequence is further filtered by path to obtain a visitor filter word grid. The customer service candidate decoding paths in the customer service filter word grid and the customer service decoding results are input into a preset rescoring model to calculate the secondary customer service score of the customer service candidate decoding paths; The visitor candidate decoding paths in the visitor filter word grid and the visitor decoding results are input into the rescoring model to calculate the visitor secondary score of the visitor candidate decoding paths.
4. The customer service voice recognition method as described in claim 1, characterized in that, The filtering of synthesized audio from the speech conversation information to be identified includes: The initial voice conversation information is processed into frames to obtain customer service voice frames; Voice endpoint detection is performed on the customer service voice frames to determine the silence segmentation point of the voice conversation information; The voice conversation information is segmented into silence based on the silence segmentation point to determine the silent voice sequence in the voice conversation information; Filter the silent speech sequences in the voice conversation information to obtain the filtered voice conversation information.
5. The customer service voice recognition method as described in claim 1, characterized in that, The filtering of synthesized audio from the speech conversation information to be identified includes: Obtain the voiceprint feature vector of the speech conversation information to be identified; The similarity between the voiceprint feature vector and the preset human voice voiceprint vector is performed to obtain the voiceprint similarity of the voiceprint feature vector; The voiceprint similarity is compared with a preset human voice matching threshold. If the voiceprint similarity is less than the human voice matching threshold, the voice conversation information is determined to be synthetic audio, and the synthetic audio is filtered out.
6. The customer service voice recognition method as described in any one of claims 1 to 5, characterized in that, The process of clustering the filtered voice conversation information to obtain customer service voice sequences and visitor voice sequences includes: Read customer service interaction information and preset the number of speakers in the voice conversation information based on the customer service interaction information; Obtain each voiceprint feature vector from the filtered voice conversation information, and parse the voiceprint feature vector to obtain the speaker information of the voiceprint feature vector; Based on the number of speakers and the speaker information, the filtered voice conversation information is clustered to obtain customer service voice sequences and visitor voice sequences.
7. A customer service voice recognition device, characterized in that, The customer service voice recognition device includes: The synthesis filtering module is configured to acquire the speech conversation information to be recognized; The initial decoding module is configured to filter the synthesized audio in the voice conversation information, and cluster the filtered voice conversation information to obtain customer service voice sequences and visitor voice sequences. The secondary scoring module is configured to decode the customer service voice sequence to obtain the customer service decoding result; and to decode the visitor voice sequence to obtain the visitor decoding result. The secondary decoding module is configured to: obtain the customer service starting voice sequence of the customer service candidate decoding path; obtain a reference customer service voice sequence associated with the customer service starting voice sequence in the customer service decoding result; decode the customer service candidate decoding path to obtain customer service decoding candidate results; obtain a reference decoding result corresponding to a preset sub-sequence of the reference customer service voice sequence; associate and concatenate the reference decoding result and the customer service decoding candidate results to obtain a concatenated candidate result; and perform a secondary scoring calculation on the concatenated candidate result to obtain a secondary customer service score for the customer service candidate decoding path. The customer service candidate decoding path is obtained by further filtering the customer service candidate decoding sequence; the customer service candidate decoding sequence is obtained by initially decoding the customer service voice sequence.
8. A customer service voice recognition device, characterized in that, The customer service voice recognition device includes: One or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the customer service voice recognition method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to execute the steps of the customer service voice recognition method according to any one of claims 1 to 6.
Citation Information
Patent Citations
Speech recognition method and device and electronic equipment
CN113362813A
Intelligent track division method, device and system for monaural call recording
CN113689861A
