A method and system for identifying roles of two-person conversation voice information

Abstract

The application provides a method and system for identifying roles of two-person dialogue voice information, the method comprising: obtaining voice information of a role to be identified; inputting the voice information of the role to be identified into a first prediction model of role identification to obtain a first role prediction result corresponding to the voice information of the role to be identified; inputting the voice information of the role to be identified into a second prediction model of role identification to obtain a second role prediction result corresponding to the voice information of the role to be identified; judging whether to adjust a first role and a second role in the first role prediction result according to the first role prediction result, the second role prediction result and a preset threshold, and determining a target role identification result of the voice information of the role to be identified from the first role prediction result according to an obtained adjustment result. The accuracy of role identification of two-person dialogue voice information is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech recognition, in particular to a method and system for role recognition of two-person dialogue speech information. BACKGROUND

[0002] With the development of the telephone customer service industry, quantification of service quality of customer service and analysis of customer demand through call content are becoming more and more important. At present, the quantification of service quality of customer service usually has real-time detection and offline detection. Real-time detection distinguishes between customer service and customers through data stream sources, but most enterprises do not deploy real-time detection, so they can only use offline detection.

[0003] Offline detection starts from recording, but in most cases, the recording does not have split data, and it is necessary to distinguish between the two roles of customer service and customers to become the key to quantifying the service quality of customer service. Only by accurately distinguishing between customer service and customers can further in-depth analysis be carried out on whether the customer service is standard, what the customer's demand is, etc.

[0004] ASR tools often also have the ability to separate roles. In general, ASR can only distinguish between two voices, but cannot correctly identify which voice is the customer service and which voice is the customer. If ASR accurately identifies the roles of customers and customer service, the accuracy rate of role recognition is 100%, and if ASR reverses the roles of customers and customer service, the accuracy rate of role recognition is 0. Therefore, it is necessary to improve the role separation ability of ASR tools. SUMMARY

[0005] Therefore, the present application provides a method and system for role recognition of two-person dialogue speech information, which solves the problem of zero accuracy rate of role recognition when the roles are reversed in the role separation of ASR tools.

[0006] The embodiment of the present application provides a method for role recognition of two-person dialogue speech information, which comprises:

[0007] Obtaining speech information of a role to be recognized, wherein the speech information of the role to be recognized is dialogue information containing a first role and a second role;

[0008] Inputting the speech information of the role to be recognized into a first prediction model of role recognition to obtain a first role prediction result corresponding to the speech information of the role to be recognized;

[0009] Inputting the speech information of the role to be recognized into a second prediction model of role recognition to obtain a second role prediction result corresponding to the speech information of the role to be recognized;

[0010] According to the role prediction first result, the role prediction second result and a preset threshold, it is judged whether the first role and the second role in the role prediction first result are adjusted, and a target role recognition result of the voice information of the to-be-recognized role is determined from the role prediction first result according to an obtained adjustment result.

[0011] Further, the obtaining of the voice information of the to-be-recognized role further comprises:

[0012] After the voice information of the to-be-recognized role is translated into voice text of the to-be-recognized role, the voice text of the to-be-recognized role is input into the first prediction model of the role recognition and the second prediction model of the role recognition.

[0013] Further, the first prediction model of the role recognition is a trained ASR role prediction model, and the second prediction model of the role recognition is a trained BERT model.

[0014] The ASR role prediction model and the BERT model are trained by the following steps, comprising:

[0015] The historical voice information is translated, the voice text after translation is role-labeled, and a role result of the voice text after translation of each historical voice information is obtained;

[0016] The voice text after translation of each historical voice information is taken as an independent variable, and the role result of the voice text after translation of each historical voice information is taken as a dependent variable, and is input into the ASR role prediction model and the BERT model for training.

[0017] Further, according to the role prediction first result, the role prediction second result and a preset threshold, it is judged whether the first role and the second role in the role prediction first result are adjusted, and a target role recognition result of the voice information of the to-be-recognized role is determined from the role prediction first result according to an obtained adjustment result, comprising:

[0018] A first probability that the first role of the role prediction first result is predicted as the second role in the role prediction second result is obtained;

[0019] A second probability that the second role of the role prediction first result is predicted as the first role in the role prediction second result is obtained;

[0020] It is judged whether the sum of the first probability and the second probability is greater than a preset threshold;

[0021] If so, the first role and the second role identified in the role prediction first result are exchanged, and the role identification result after the exchange is a target role identification result; otherwise, the role prediction first result is taken as the target role identification result.

[0022] Further, before inputting the translated speech text of the historical speech information into the ASR role prediction model and the BERT model, the translated speech text of the historical speech information is preprocessed.

[0023] The preprocessing includes disordering the conversation order of all the translated speech texts of the historical speech information, and merging the disordered speech texts.

[0024] The embodiment of the application also provides a role identification system for two-person dialogue speech information, which comprises:

[0025] An information acquisition module is configured to acquire speech information of a role to be identified, wherein the speech information of the role to be identified is dialogue information containing a first role and a second role.

[0026] A first identification module is configured to input the speech information of the role to be identified into a first prediction model for role identification, to obtain a role prediction first result.

[0027] A second identification module is configured to input the speech information of the role to be identified into a second prediction model for role identification, to obtain a role prediction second result corresponding to the speech information of the role to be identified.

[0028] A role identification adjustment module is configured to determine whether to adjust the first role and the second role in the role prediction first result according to the role prediction first result, the role prediction second result and a preset threshold, and determine a target role identification result of the speech information of the role to be identified from the role prediction first result according to the obtained adjustment result.

[0029] Further, the system further comprises:

[0030] A translation module is configured to translate the speech information of the role to be identified into speech text of the role to be identified, and then input the speech text of the role to be identified into the first prediction model for role identification and the second prediction model for role identification.

[0031] Further, the first prediction model for role identification is a trained ASR role prediction model, and the second prediction model for role identification is a trained BERT model; and

[0032] The system further comprises a training module configured to acquire the trained ASR role prediction model and the BERT model, and the training module comprises:

[0033] A marking unit is configured to translate each historical voice information, mark the translated voice text with a role, and obtain a role result of the translated voice text of each historical voice information;

[0034] A training unit is configured to input the translated voice text of each historical voice information as an independent variable and the role result of the translated voice text of each historical voice information as a dependent variable into the ASR role prediction model and the BERT model for training.

[0035] Further, the role recognition adjustment module comprises:

[0036] A first probability obtaining unit is configured to obtain a first probability that the first role in the first role prediction result is predicted as the second role in the second role prediction result.

[0037] A second probability obtaining unit is configured to obtain a second probability that the second role in the first role prediction result is predicted as the first role in the second role prediction result.

[0038] A judging unit is configured to judge whether the sum of the first probability and the second probability is greater than a preset threshold.

[0039] A target role obtaining unit is configured to, when the sum of the first probability and the second probability is greater than the preset threshold, exchange the first role and the second role recognized in the first role prediction result, and the exchanged role recognition result is a target role recognition result; otherwise, the first role prediction result is taken as the target role recognition result.

[0040] Further, the system further comprises a preprocessing module configured to preprocess the translated voice text of the historical voice information before the translated voice text is input into the ASR role prediction model and the BERT model.

[0041] The preprocessing module comprises:

[0042] A disordering module is configured to disorder the conversation order of all the translated voice text of the historical voice information.

[0043] A merging module is configured to merge the disordered voice text.

[0044] The application provides a role recognition method for two-person dialogue voice information, which comprises the following steps: inputting voice information of a to-be-recognized role into a first prediction model for role recognition to obtain a role prediction first result corresponding to the voice information of the to-be-recognized role; inputting the voice information of the to-be-recognized role into a second prediction model for role recognition to obtain a role prediction second result corresponding to the voice information of the to-be-recognized role; judging whether to adjust a first role and a second role in the role prediction first result according to the role prediction first result, the role prediction second result and a preset threshold, and determining a target role recognition result of the voice information of the to-be-recognized role from the role prediction first result according to an obtained adjustment result. The role recognition accuracy of two-person dialogue voice information is improved, and the problem that the role recognition accuracy is zero when the roles are opposite in the role separation of an ASR tool is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0046] Figure 1 A flowchart of a role recognition method for two-person dialogue voice information provided by the embodiments of the present application;

[0047] Figure 2 A flowchart of adjusting a target role recognition result of voice information of a to-be-recognized role provided by the embodiments of the present application;

[0048] Figure 3 A functional module diagram of a first role recognition system for two-person dialogue voice information provided by the embodiments of the present application;

[0049] Figure 4 A functional module diagram of a second role recognition system for two-person dialogue voice information provided by the embodiments of the present application;

[0050] Figure 5 A functional module diagram of a third role recognition system for two-person dialogue voice information provided by the embodiments of the present application;

[0051] Figure 6 A functional module diagram of a fourth role recognition system for two-person dialogue voice information provided by the embodiments of the present application;

[0052] Figure 7 A functional module diagram of a fifth role recognition system for two-person dialogue voice information provided by the embodiments of the present application. DETAILED DESCRIPTION

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting upon the application; the terms "comprising," "including," and "having," and variations thereof, as used in enrolling and claims herein, are intended to be open-ended terms that specify the presence of the stated elements but do not preclude the presence of additional elements. The terms "first," "second," and the like, as used herein do not imply a serial or chronological order, unless otherwise indicated.

[0054] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase that the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of other embodiments. It is explicitly contemplated that embodiments described herein can be combined with other embodiments.

[0055] In order to make the purposes, technical solutions and advantages of the application more clear, the application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0056] In the following description, the suffix used for an element such as "module", "part", or "unit" is only for convenience of explanation of the application, and does not have a specific meaning by itself.

[0057] The scheme provided by the embodiments of the application relates to artificial intelligence deep learning (DL), voice technology and the like.

[0058] Artificial Intelligence (AI) is the theory, method, technology and application system of using digital computer or digital computer controlled machine to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc.

[0059] Deep Learning (DL) is a major research direction in the field of Machine Learning (ML), which is introduced into machine learning to make it closer to the original goal-artificial intelligence. Deep learning is to learn the internal rules and representation levels of sample data, and the information obtained in the learning process is very helpful for the interpretation of data such as text, images and sound. The ultimate goal is to enable machines to have analysis and learning ability like people, and to recognize text, images and sound data. Deep learning is a complex machine learning algorithm, which has achieved much better results in speech and image recognition than previous related technologies. Deep learning has achieved many results in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technology, and other related fields. Deep learning enables machines to imitate human activities such as vision, hearing and thinking, solves many complex pattern recognition problems, and makes great progress in artificial intelligence related technologies.

[0060] The key technologies of speech technology include automatic speech recognition technology (ASR), speech synthesis technology (TTS) and voiceprint recognition technology. Let the computer can hear, can see, can say, can feel, is the development direction of future human-computer interaction, among which speech has become one of the most promising human-computer interaction methods in the future.

[0061] Figure 1is a flowchart of a method for role recognition of two-person dialogue voice information provided by an embodiment of the present application. The present application provides method operation steps as described in the embodiments or flowcharts, but more or fewer operation steps can be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is only one of the many execution orders of the steps, and does not represent the only execution order. In actual system or server product execution, the method order shown in the embodiments or the drawings can be executed in sequence or in parallel (for example, in a parallel processor or multi-thread processing environment). Please refer to Figure 1 The method for role recognition of two-person dialogue voice information provided by an embodiment of the present application can include the following steps:

[0062] Step S1: Obtain voice information of a role to be recognized, wherein the voice information of the role to be recognized is dialogue information containing a first role and a second role;

[0063] Voice information of a role to be recognized is obtained, wherein voice information refers to voice data generated in a dialogue process between two or more people. After obtaining the voice information, the roles of the voice information need to be distinguished, and the dialogue content is detected according to the distinguished roles to realize tracking analysis of the roles. In some actual scenarios, for example, some consulting services, there will be a dialogue between a customer and a customer service. Voice customer service mainly includes customer service in the form of telephone or online voice. Voice data can be generated in the process of telephone or voice dialogue, and the roles of these dialogue data need to be recognized. According to the recognized roles, the dialogue content between the customer and the customer service is analyzed in depth, and then it is detected whether the customer service is speaking in a standard manner and what the actual needs of the customer are.

[0064] Step S2: Input the voice information of the role to be recognized into a first prediction model for role recognition to obtain a first role prediction result corresponding to the voice information of the role to be recognized;

[0065] In an embodiment of the present application, the first prediction model for role recognition is an ASR role prediction model. Automatic speech recognition is also called speech recognition or computer speech recognition. The voice feature parameters of the role are analyzed in advance, a voice template is made, and is stored in a voice parameter library. The voice information to be recognized is analyzed in the same way as during training to obtain voice parameters. It is compared with the reference template in the library one by one, and a decision method is used to find the template closest to the voice feature to obtain the role recognition result of the voice information.

[0066] In the embodiments of the present application, the role prediction first result can include a dialogue between the first role and the second role identified according to the voice information feature parameter feature information, for example, some consulting services, inputting the dialogue between the customer and the customer service into automatic speech recognition (ASR), and the automatic speech recognition (ASR) identifies the dialogue as the dialogue between the customer (the first role) and the customer service (the second role).

[0067] Step S3: inputting the voice information of the role to be identified into a second prediction model of role identification to obtain a role prediction second result corresponding to the voice information of the role to be identified;

[0068] In the embodiments of the present application, the second prediction model of role identification can be a trained BERT model. The network architecture of BERT is a multi-layer bidirectional Transformer encoder based on the original implementation described in Vaswani et al. (2017). BERT uses a bidirectional Transformer to learn context feature information in all layers at the same time. The Bert model uses a self-supervised learning method to pre-train on a large amount of unlabeled corpus, uses a Transformer Encoder structure, and converts the distance between two words at any position into a vector representation through an Attention mechanism, effectively solves the problem of long-term dependence in natural language processing, and at the same time obtains rich semantic information in the text.

[0069] In the embodiments of the present application, the role prediction second result can include a dialogue between the first role and the second role identified according to the voice information semantic information, for example, some consulting services, which are identified as the dialogue between the customer (the first role) and the customer service (the second role) by the BERT model.

[0070] In some embodiments, before step S2 and step S3, the ASR role prediction model and the BERT model need to be trained first, and the parameters of the ASR role prediction model and the BERT model are determined.

[0071] Specifically, before training the ASR role prediction model and the BERT model, the historical voice information stored in the database is first translated into voice text, and the voice text after translation is role-labeled to obtain the role result of the voice text after translation of each historical voice information. For example, it can be converted into voice text by automatic speech recognition (ASR), and then the voice text after translation is role-labeled to obtain the role result of the voice text after translation of each historical voice information. It needs to be explained that the category is the first role and the second role contained in the voice information, and the role labeling refers to the process of establishing a mapping relationship between the voice text after translation and the corresponding role. For example, some consulting services mark the words of the customer in the voice text as 1, and mark the words of the customer service in the voice text as 0.

[0072] Further, the voice text after translation of each historical voice information is input into the ASR role prediction model and the BERT model as an independent variable, and the role result of the voice text after translation of each historical voice information is input into the ASR role prediction model and the BERT model as a dependent variable. Then, the words of the customer and the customer service in the voice text after translation of each historical voice information are input into the ASR role prediction model and the BERT model as independent variables, and the words of the customer in the voice text are marked as 1 and the words of the customer service in the voice text are marked as 0 as dependent variables. Input into the ASR role prediction model and the BERT model for training. After training, the ASR role prediction model and the BERT model obtained by training are verified, and when the accuracy rate reaches 80% or more, the ASR role prediction model and the BERT model obtained by training are used as the first prediction model of the role recognition and the second prediction model of the role recognition.

[0073] In some embodiments, the voice text after translation of the historical voice information is input into the ASR role prediction model and the BERT model before the voice text after translation of the historical voice information is input into the ASR role prediction model and the BERT model.

[0074] The preprocessing includes disordering the conversation order of all the voice text after translation of the historical voice information, and merging the voice text after disordering.

[0075] The historical voice information is voice data generated in the conversation process between two or more people, and the voice text is obtained by translation. In order to improve the prediction accuracy of the ASR role prediction model and the BERT model, the conversation order of all the voice text after translation of the historical voice information is disordered, and then the voice text after disordering is randomly merged before being input into the ASR role prediction model and the BERT model for training.

[0076] Step S4: judging whether to adjust the first role and the second role in the role prediction first result according to the role prediction first result, the role prediction second result and a preset threshold, and determining a target role recognition result of the voice information of the to-be-identified role from the role prediction first result according to an obtained adjustment result.

[0077] From the above embodiment, it can be known that the first prediction model of the role recognition identifies each call voice information as a dialogue between the first role and the second role, the second prediction model of the role recognition identifies each call voice information as a dialogue between the first role and the second role, then the role prediction first result and the role prediction second result are compared, a probability that the first prediction model identifies voice information as the first role and the second prediction model identifies the same voice information as the second role can be obtained, and similarly, a probability that the first prediction model identifies voice information as the second role and the second prediction model identifies the same voice information as the first role can be obtained, then the two probabilities are added and compared with a preset threshold, if the sum of the two probabilities exceeds the preset threshold, the role of the voice information corresponding to the first role identified in the first prediction model is replaced with the second role, and the role of the voice information corresponding to the second role identified in the first prediction model is replaced with the first role, and then the replaced result is taken as the target role recognition result of the voice information of the to-be-identified role. If the sum of the two probabilities is within the preset threshold, the first role and the second role of the voice information of the to-be-identified role identified by the first prediction model of the role recognition are directly taken as the target role recognition result.

[0078] Figure 2 is a flowchart of adjusting a target role recognition result of voice information of a to-be-identified role provided by an embodiment of the present application, specifically, step S4 includes:

[0079] Step S41: obtaining a first probability that a first role of the role prediction first result is predicted as a second role in the role prediction second result;

[0080] Step S42: obtaining a second probability that a second role of the role prediction first result is predicted as a first role in the role prediction second result;

[0081] Step S43: judging whether a sum of the first probability and the second probability is greater than a preset threshold;

[0082] Step S44: if yes, the positions of the first role and the second role identified in the role prediction first result are exchanged, and an exchanged role recognition result is taken as the target role recognition result; otherwise, the role prediction first result is taken as the target role recognition result.

[0083] Specifically, for example, some consulting services input the conversation between the customer and the customer service into an automatic speech recognition (ASR) and a BERT model, the automatic speech recognition (ASR) recognizes the conversation as the role of the customer and the customer service, the BERT model recognizes the conversation as the role of the customer and the customer service, then the first probability of the same conversation content being recognized as the customer by the automatic speech recognition (ASR) but as the customer service by the BERT model is counted, and the second probability of the same conversation content being recognized as the customer service by the automatic speech recognition (ASR) but as the customer by the BERT model is counted, then the first probability and the second probability are added, compared with a preset threshold, if the sum of the first probability and the second probability exceeds the preset threshold, the voice information of the to-be-identified role recognized by the automatic speech recognition (ASR) is recognized as the customer's conversation role and replaced with the customer service, the voice information of the to-be-identified role recognized by the automatic speech recognition (ASR) is recognized as the customer service's conversation role and replaced with the customer, and the result after the replacement is the target role recognition result of the voice information of the to-be-identified role. If the sum of the first probability and the second probability is within the preset threshold, the role result of the automatic speech recognition (ASR) recognizing the conversation as the role of the customer and the customer service is directly taken as the target role recognition result of the voice information of the to-be-identified role.

[0084] The application provides a role recognition method for two-person conversation voice information, which comprises the following steps: inputting voice information of a to-be-identified role into a first prediction model for role recognition to obtain a role prediction first result corresponding to the voice information of the to-be-identified role; inputting the voice information of the to-be-identified role into a second prediction model for role recognition to obtain a role prediction second result corresponding to the voice information of the to-be-identified role; and judging whether to adjust a first role and a second role in the role prediction first result according to the role prediction first result, the role prediction second result and a preset threshold, and determining a target role recognition result of the voice information of the to-be-identified role from the role prediction first result according to the obtained adjustment result. The accuracy of role recognition of two-person conversation voice information is improved, and the problem of zero accuracy of role recognition when the roles are opposite in the role separation of an ASR tool is avoided.

[0085] Figure 3A function module schematic diagram of a two-person dialogue voice information role recognition system provided by an embodiment of the present application, the system comprises: an information acquisition module 1, used for acquiring voice information of a to-be-recognized role, wherein the voice information of the to-be-recognized role is dialogue information containing a first role and a second role; a first recognition module 2, used for inputting the voice information of the to-be-recognized role into a first prediction model of role recognition to obtain a role prediction first result; a second recognition module 3, used for inputting the voice information of the to-be-recognized role into a second prediction model of role recognition to obtain a role prediction second result corresponding to the voice information of the to-be-recognized role; a role recognition adjustment module 4, used for judging whether to adjust the first role and the second role in the role prediction first result according to the role prediction first result, the role prediction second result and a preset threshold, and determining a target role recognition result of the voice information of the to-be-recognized role from the role prediction first result according to an obtained adjustment result.

[0086] On the basis of the above-mentioned embodiments, in other embodiments, Figure 4 A function module schematic diagram of a two-person dialogue voice information role recognition system provided by an embodiment of the present application, the system further comprises: a translation module 5, used for translating the voice information of the to-be-recognized role into voice text of the to-be-recognized role, and then inputting the voice text of the to-be-recognized role into the first prediction model of role recognition and the second prediction model of role recognition.

[0087] On the basis of the above-mentioned embodiments, in other embodiments, Figure 5 A function module schematic diagram of a two-person dialogue voice information role recognition system provided by an embodiment of the present application, the system further comprises a training module 6, used for acquiring a trained ASR role prediction model and a BERT model, the training module comprises: a marking unit 61, used for translating each historical voice information, marking the roles of the translated voice text, and acquiring a role result of the translated voice text of each historical voice information; a training unit 62, used for inputting the translated voice text of each historical voice information as an independent variable and the role result of the translated voice text of each historical voice information as a dependent variable into the ASR role prediction model and the BERT model for training.

[0088] On the basis of the above-mentioned embodiments, in other embodiments, Figure 6A function module schematic diagram of a two-person dialogue voice information role recognition system provided by an embodiment of the present application is provided, and the role recognition adjustment module 4 comprises: a first probability acquisition unit 41, configured to acquire a first probability that a first role in a first role prediction result is predicted as a second role in a second role prediction result; a second probability acquisition unit 42, configured to acquire a second probability that a second role in the first role prediction result is predicted as the first role in the second role prediction result; a judgment unit 43, configured to judge whether the sum of the first probability and the second probability is greater than a preset threshold; and a target role acquisition unit 44, configured to, when the sum of the first probability and the second probability is greater than the preset threshold, exchange the first role and the second role recognized in the first role prediction result, and the exchanged role recognition result is a target role recognition result; otherwise, the first role prediction result is taken as the target role recognition result.

[0089] On the basis of the above-mentioned embodiments, in other embodiments, Figure 7 A function module schematic diagram of a two-person dialogue voice information role recognition system provided by an embodiment of the present application is provided, and the system further comprises a preprocessing module 7, configured to perform preprocessing on the voice text after historical voice information is translated before the voice text after the historical voice information is translated is input into an ASR role prediction model and a BERT model; the preprocessing module 7 comprises: a disordering module 71, configured to disorder the dialogue order of all the voice text after the historical voice information is translated; and a merging module 72, configured to merge the voice text after being disordered.

[0090] For other details of the implementation of each module in the two-person dialogue voice information role recognition system provided by the above-mentioned embodiments, refer to the description in the two-person dialogue voice information role recognition method in the above-mentioned embodiments, which will not be repeated here.

[0091] It should be noted that each embodiment in the present specification adopts a progressive description manner, and each embodiment focuses on the difference from other embodiments, and the same and similar parts between embodiments can be referred to. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the description of the method embodiments.

Claims

1. A method for recognizing roles of voices in a two-person conversation, characterized by, The method comprises: obtaining voice information of a role to be identified, wherein the voice information of the role to be identified comprises dialogue information of a first role and a second role; inputting the voice information of the role to be identified into a first prediction model of role identification to obtain a first role prediction result corresponding to the voice information of the role to be identified; inputting the voice information of the role to be identified into a second prediction model of role identification to obtain a second role prediction result corresponding to the voice information of the role to be identified; judging whether to adjust the first role and the second role in the first role prediction result according to the first role prediction result, the second role prediction result and a preset threshold, and determining a target role identification result of the voice information of the role to be identified from the first role prediction result according to an obtained adjustment result; the judging whether to adjust the first role and the second role in the first role prediction result according to the first role prediction result, the second role prediction result and a preset threshold, and determining a target role identification result of the voice information of the role to be identified from the first role prediction result according to an obtained adjustment result comprises: obtaining a first probability that the first role in the first role prediction result is predicted as the second role in the second role prediction result; obtaining a second probability that the second role in the first role prediction result is predicted as the first role in the second role prediction result; judging whether a sum of the first probability and the second probability is greater than the preset threshold; if yes, exchanging the first role and the second role identified in the first role prediction result, and the exchanged role identification result is the target role identification result; otherwise, taking the first role prediction result as the target role identification result.

2. The method of claim 1, wherein the method further comprises: after the voice information of the role to be identified is obtained, the method further comprises: after the voice information of the role to be identified is translated into voice text of the role to be identified, inputting the voice text of the role to be identified into the first prediction model of role identification and the second prediction model of role identification.

3. The method of claim 2, wherein the method further comprises: the first prediction model of role identification is a trained ASR role prediction model, and the second prediction model of role identification is a trained BERT model; and the ASR role prediction model and the BERT model are trained by the following steps, comprising: translating each historical voice information, role marking the voice text translated after role marking, and obtaining a role result of the voice text translated after role marking of each historical voice information; inputting the voice text translated after role marking of each historical voice information as an independent variable and the role result of the voice text translated after role marking of each historical voice information as a dependent variable into the ASR role prediction model and the BERT model for training.

4. The method of claim 3, wherein the method further comprises: before the voice text translated after role marking of historical voice information is input into the ASR role prediction model and the BERT model, the voice text translated after role marking of the historical voice information is preprocessed; the preprocessing comprises disordering dialogue order of all the voice text translated after role marking of the historical voice information, and merging the voice text after disordering.

5. A system for identifying the roles of two people in a conversation, characterized in that the system comprises: The information acquisition module is configured to acquire voice information of a to-be-identified role, wherein the voice information of the to-be-identified role is dialogue information containing a first role and a second role. The first identification module is configured to input the voice information of the to-be-identified role into a first prediction model of role identification to obtain a role prediction first result. The second identification module is configured to input the voice information of the to-be-identified role into a second prediction model of role identification to obtain a role prediction second result corresponding to the voice information of the to-be-identified role. The role identification adjustment module is configured to determine, according to the role prediction first result, the role prediction second result, and a preset threshold, whether to adjust the first role and the second role in the role prediction first result, and determine a target role identification result of the voice information of the to-be-identified role from the role prediction first result according to an obtained adjustment result. The role identification adjustment module includes: The first probability acquisition unit is configured to acquire a first probability that the first role in the role prediction first result is predicted as the second role in the role prediction second result. The second probability acquisition unit is configured to acquire a second probability that the second role in the role prediction first result is predicted as the first role in the role prediction second result. The determination unit is configured to determine whether a sum of the first probability and the second probability is greater than the preset threshold. The target role acquisition unit is configured to, when the sum of the first probability and the second probability is greater than the preset threshold, exchange the first role and the second role identified in the role prediction first result, and the exchanged role identification result is a target role identification result; otherwise, the role prediction first result is taken as the target role identification result.

6. The system for identifying a character of a voice information of a two-person conversation according to claim 5, wherein The system further includes: The translation module is configured to translate the voice information of the to-be-identified role into voice text of the to-be-identified role, and then input the voice text of the to-be-identified role into the first prediction model of role identification and the second prediction model of role identification.

7. The system for identifying a character of a voice information of a two-person conversation according to claim 6, wherein The first prediction model of role identification is a trained ASR role prediction model, and the second prediction model of role identification is a trained BERT model; and The system further includes a training module configured to acquire the trained ASR role prediction model and the BERT model, and the training module includes: The marking unit is configured to translate each historical voice information, mark roles of the translated voice text, and acquire role results of the translated voice text of each historical voice information. The training unit is configured to input the translated voice text of each historical voice information as an independent variable and the role results of the translated voice text of each historical voice information as a dependent variable into the ASR role prediction model and the BERT model for training.

8. The system for identifying a character of a voice information of a two-person conversation according to claim 7, wherein, The system further includes a preprocessing module configured to preprocess the translated voice text of historical voice information before inputting the translated voice text of the historical voice information into the ASR role prediction model and the BERT model. The preprocessing module includes: The disordering module is configured to disorder dialogue orders of all the translated voice text of the historical voice information. A merging module is configured to merge the scrambled speech text.

Images