A method and device for audio role recognition of online car-hailing based on voiceprint clustering

Abstract

The present application relates to a method and device for audio role recognition of online ride-hailing vehicles based on voiceprint clustering, including: obtaining the original audio data recorded by the online ride-hailing vehicle during the day; extracting the human voice audio data from the original audio data; segmenting the human voice audio data to obtain multiple human voice audio data segments; clustering the human voice audio data segments to obtain a driver audio data set to be screened and a passenger audio data set; inputting the driver audio data set to be screened into a pre-trained overlap detection model, and using the overlap detection model to split the overlapping audio and the driver audio data set in the driver audio data set to be screened. In this technical solution, the driver audio data set to be screened is input into a pre-trained overlap detection model to split the overlapping audio and the driver audio data set, and finally obtain accurate driver audio data sets and passenger audio data sets, thereby solving the problem of inaccurate classification results due to factors such as voice overlap in the prior art.

Description

Technical Field

[0001] The present application relates to the field of audio recognition technology, and in particular to a method and device for audio role recognition of online car-hailing vehicles based on voiceprint clustering. Background Art

[0002] In online car-hailing services, an important measure has been taken to ensure the safety of passengers and improve the service quality of drivers: after passengers get on the car and sit down, they will immediately receive a reminder to record the entire journey, and the recording in the car will begin at this time. This measure is aimed at ensuring the transparency and safety of the journey and providing passengers with a more secure and comfortable travel experience. In every service process, it is crucial to ensure that the audio data of the driver and the passenger are fully and accurately separated. Such separation operations help to build a clear conversation sequence, and then conduct in-depth analysis of the conversation between the driver and the passenger. Through this analysis, it is possible to accurately identify possible violations of the driver during the service process, thereby providing users with a better quality and compliant travel experience.

[0003] In the existing technology, there is a problem of unclear classification, especially when separating the voices of drivers and passengers, it is difficult to accurately determine whether the voice is from the driver, passenger or customer service. Since different people's voiceprints have unique characteristics, these features can be used to distinguish different speakers. Traditional audio feature extraction methods such as MFCC, PLP, Deep Feature, etc., these features can reflect the voice characteristics of different people. However, there is a problem of inaccurate classification results due to factors such as voice overlap. Summary of the invention

[0004] In order to at least overcome to some extent the problem in the related art of inaccurate classification results due to factors such as voice overlap when performing audio role recognition on audio data recorded in an online car-hailing vehicle, the present application provides an online car-hailing audio role recognition method and device based on voiceprint clustering.

[0005] The scheme of this application is as follows:

[0006] According to a first aspect of an embodiment of the present application, a method for audio role recognition of an online car-hailing service based on voiceprint clustering is provided, comprising:

[0007] Obtain the original audio data recorded by online ride-hailing vehicles during the day;

[0008] Extracting human voice audio data from the original audio data;

[0009] Segmenting the human voice audio data to obtain a plurality of human voice audio data segments;

[0010] Clustering the human voice audio data segments to obtain a driver audio data set and a passenger audio data set to be screened;

[0011] The driver audio data set to be screened is input into a pre-trained overlap detection model, and the overlap detection model is used to separate the overlapping audio and the driver audio data set from the driver audio data set to be screened.

[0012] Preferably, clustering the human voice audio data segments to obtain the driver audio data set and the passenger audio data set to be screened includes:

[0013] Determine the cosine similarity between two consecutive human voice audio data segments, and merge them if the cosine similarity between the two consecutive human voice audio data segments is higher than a first preset threshold;

[0014] When the cosine similarity between any two consecutive human voice audio data segments is not higher than a first preset threshold, clustering the current audio data segments, and dividing the clusters longer than the first audio duration into a first cluster data set, and dividing the clusters not longer than the first audio duration into a second cluster data set;

[0015] Take the average of all embedding vectors in each cluster dataset to get the central embedding vector;

[0016] According to the central embedding vector of each cluster data set, the cosine similarity between the cluster data sets is determined;

[0017] When the cosine similarity between the second type of clustering data set and any first type of clustering data set is higher than a second preset threshold, merge them into the first type of clustering data set with the highest cosine similarity;

[0018] When the cosine similarity between the second type of clustered data set and any first type of clustered data set is not higher than a second preset threshold, determining it as a passenger audio data set;

[0019] The non-passenger audio data set is determined as the driver audio data set to be screened.

[0020] Preferably, the method further comprises:

[0021] Obtain a driver audio sample dataset and an overlapping audio sample dataset;

[0022] The overlap detection model is trained by using a plurality of driver audio sample data sets and an overlap audio sample data set as training data sets.

[0023] Preferably, obtaining a driver audio sample data set includes:

[0024] The driver audio data collected when the driver registers and the historical driver audio data set are obtained as the driver audio sample data set.

[0025] Preferably, the method further comprises:

[0026] The driver audio sample data set is standardized; including:

[0027] Segmenting the driver audio sample data in the driver audio sample data set;

[0028] Performing Mel-spectrogram conversion on the segmented driver audio sample data;

[0029] Calculate the mean and standard deviation of the driver audio sample data after the Mel-spectrogram conversion and perform normalization.

[0030] Preferably, training the overlap detection model comprises:

[0031] Build an overlap detection model;

[0032] Inputting the training data set into an overlap detection model, and iterating based on a contrastive learning loss function and a cross entropy loss function;

[0033] After the iteration is completed, the overlap detection model is evaluated, and the training is completed after the evaluation passes.

[0034] Preferably, segmenting the human voice audio data comprises:

[0035] The first part of the human voice audio data is divided based on the second audio duration, and the remaining part is divided based on the third audio duration;

[0036] The driver audio sample data is segmented, including:

[0037] The first part of the audio data in the driver audio sample data set is divided based on the second audio duration, and the remaining part is divided based on the third audio duration;

[0038] The duration of the second audio is four times the duration of the third audio.

[0039] Preferably, the method further comprises:

[0040] At every preset period, the overlap detection model is updated using the driver audio data set within the period.

[0041] Preferably, the method further comprises:

[0042] The driver audio data set is stored and marked with the corresponding date.

[0043] According to a second aspect of an embodiment of the present application, a device for identifying audio roles of online car-hailing vehicles based on voiceprint clustering is provided, comprising:

[0044] Processor and memory;

[0045] The processor and the memory are connected via a communication bus:

[0046] Wherein, the processor is used to call and execute the program stored in the memory;

[0047] The memory is used to store a program, and the program is at least used to execute a method for audio role recognition of an online car-hailing vehicle based on voiceprint clustering as described in any one of the above items.

[0048] The technical solution provided by this application may have the following beneficial effects:

[0049] The online car-hailing audio role recognition method based on voiceprint clustering in the present application includes: obtaining the original audio data recorded by the online car-hailing during the day; extracting the human voice audio data in the original audio data; segmenting the human voice audio data to obtain multiple human voice audio data segments; clustering the human voice audio data segments to obtain a driver audio data set to be screened and a passenger audio data set; inputting the driver audio data set to be screened into a pre-trained overlap detection model, and splitting the overlapping audio and the driver audio data set in the driver audio data set to be screened through the overlap detection model.

[0050] The technical solution in this application first divides the human voice audio data in the original audio data recorded by the online car-hailing service during the day and then clusters them. Since the driver of the online car-hailing service is a permanent role and the passenger is a rotating role, the clustered data set generally contains a larger driver audio data set and multiple smaller passenger audio data sets. Since the driver audio data set to be screened may be interfered by overlapping audio, in this technical solution, the driver audio data set to be screened is also input into a pre-trained overlap detection model to split the overlapping audio and the driver audio data set, and finally obtain accurate driver audio data sets and passenger audio data sets, thereby solving the problem of inaccurate classification results due to factors such as voice overlap in the prior art.

[0051] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0053] Figure 1 It is a flowchart of a method for audio role recognition of online car-hailing based on voiceprint clustering provided by an embodiment of the present application;

[0054] Figure 2It is a schematic diagram of a process of clustering human voice audio data segments in a method for audio role recognition of online car-hailing based on voiceprint clustering provided by an embodiment of the present application;

[0055] Figure 3 It is a structural diagram of an online car-hailing audio role recognition device based on voiceprint clustering provided by an embodiment of the present application.

[0056] Reference numerals: processor-21; memory-22. DETAILED DESCRIPTION

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0058] Embodiment 1

[0059] Figure 1 This is a flow chart of a method for audio role recognition of online car-hailing based on voiceprint clustering provided by an embodiment of the present application, with reference to Figure 1 , a method for audio role recognition of online car-hailing based on voiceprint clustering, comprising:

[0060] S1: Obtain the original audio data recorded by the online ride-hailing vehicle during the day;

[0061] Online taxi, short for online taxi booking service, refers to the business activities of building a service platform based on Internet technology, integrating supply and demand information, using qualified vehicles and drivers, and providing non-cruising taxi booking services. Users can select the model and driver through mobile phone applications, quickly call a taxi and enjoy travel services. Compared with traditional taxis, online taxis have the advantages of transparent prices, convenient calling, and controllable service quality, so they are rapidly becoming popular around the world.

[0062] In online car-hailing services, an important measure has been taken to ensure the safety of passengers and improve the service quality of drivers: after the passengers get on the car and sit down, they will immediately receive a reminder to record the entire journey, and the recording inside the car will begin at this time.

[0063] Intraday means within one day. In the current day, in this embodiment, the original audio data recorded by the online car-hailing service is obtained on a daily basis, so as to monitor the driver's daily behavior and improve real-time performance.

[0064] The original audio data refers to all the audio data from the time the passenger gets on the bus and sits down to the time the passenger gets off the bus. Multiple segments of the above original audio data will be generated in a day.

[0065] S2: extracting human voice audio data from the original audio data;

[0066] It should be noted that since passengers generally do not talk to the driver all the time during the ride, there will be a large amount of unvoiced audio in the original audio data. This part of the unvoiced audio needs to be discarded, leaving only the human voice audio data for subsequent clustering.

[0067] S3: Segment the human voice audio data to obtain multiple human voice audio data segments;

[0068] Segment the human voice audio data, including:

[0069] The first part of the human voice audio data is divided based on the second audio duration, and the remaining part is divided based on the third audio duration;

[0070] The duration of the second audio is four times the duration of the third audio.

[0071] For example:

[0072] For a segment of human voice audio data, the first segmentation is performed after an interval of 1280 milliseconds at the beginning, and then a new segmentation is performed each time the segmentation slides 320 milliseconds.

[0073] S4: clustering the human voice audio data segments to obtain the driver audio data set and the passenger audio data set to be screened;

[0074] Reference Figure 2 , specifically including:

[0075] S41: Determine the cosine similarity between two consecutive human voice audio data segments, and if the cosine similarity between the two consecutive human voice audio data segments is higher than a first preset threshold, merge them;

[0076] Preferably, in this embodiment, the first preset threshold is set to 0.7. If the cosine similarity between two consecutive human voice audio data segments is higher than 0.7, the two consecutive human voice audio data segments are regarded as audio data emitted by the same person, so they are merged.

[0077] S42: When the cosine similarity between any two consecutive human voice audio data segments is not higher than a first preset threshold, clustering the current audio data segments, and dividing the clusters longer than the first audio duration into a first cluster data set, and dividing the clusters not longer than the first audio duration into a second cluster data set;

[0078] When the cosine similarity between any two consecutive human voice audio data segments is not higher than the first preset threshold, it is deemed that all the human voice audio data segments have been merged, and any two consecutive human voice audio data segments currently no longer meet the conditions for merging. At this time, the status of the human voice audio data segments should be: driver audio data segment, passenger audio data segment, driver audio data segment, passenger audio data segment, and so on.

[0079] Preferably, in this embodiment, the first audio duration is set to 6s, the current audio data segment is clustered, and the clusters with a duration longer than 6s are divided into the first type of cluster data set (long cluster), and the clusters with a duration not longer than 6s are divided into the second type of cluster data set (short cluster), and the dividing line between long and short clusters is 6 seconds.

[0080] S43: Take the average of all embedding vectors in each cluster dataset to obtain the central embedding vector;

[0081] In this embodiment, the acoustic features of each clustered data set can be extracted as its embedding vector by encoding in the pre-training mode of SpeechBrain. SpeechBrain is a speech processing toolkit from many research institutions such as the University of Montreal and Yoshua Bengio. Users can quickly implement tasks such as speech recognition, voiceprint recognition, speech enhancement, signal processing, etc., which will not be described in detail here.

[0082] Then, we take the average of all embedding vectors in each cluster dataset to get the central embedding vector as the cluster center.

[0083] S44: judging the cosine similarity between the cluster data sets according to the central embedding vector of each cluster data set;

[0084] Here, the cosine similarity between each second-class clustering data set and the first-class clustering data set is mainly determined.

[0085] S45: When the cosine similarity between the second clustering data set and any first clustering data set is higher than a second preset threshold, merge them into the first clustering data set with the highest cosine similarity;

[0086] Preferably, in this embodiment, the second preset threshold is set to 0.2. When the cosine similarity between the second clustering data set and any first clustering data set is higher than 0.2, it is merged into the first clustering data set with the highest cosine similarity.

[0087] For example, if the cosine similarities of the current second-class clustering data set with the first-class clustering data set A, the first-class clustering data set B, and the first-class clustering data set C are 0.6, 0.7, and 0.8 respectively, the current second-class clustering data set is merged into the first-class clustering data set C.

[0088] S46: When the cosine similarity between the second type of clustered data set and any first type of clustered data set is not higher than a second preset threshold, determining it as a passenger audio data set;

[0089] When the cosine similarity between the second-category clustering dataset and any first-category clustering dataset is not higher than 0.2, it is considered that the second-category clustering dataset represents a new speaker, namely, the passenger, and is therefore determined as a passenger audio dataset.

[0090] S47: Determine the non-passenger audio data set as the driver audio data set to be screened.

[0091] Since the drivers of online ride-hailing cars are permanent roles and the passengers are rotating roles, the clustered dataset generally contains a larger driver audio dataset and multiple smaller passenger audio datasets.

[0092] In this embodiment, after excluding these passenger audio data sets, the non-passenger audio data sets are first classified as the driver audio data set to be screened, and the driver audio data set to be screened generally includes the driver audio data set, noise, and overlapping audio.

[0093] S5: Input the driver audio dataset to be screened into a pre-trained overlap detection model, and use the overlap detection model to split the driver audio dataset to be screened into overlapping audio and driver audio datasets.

[0094] Since the driver audio data set to be screened may be interfered by overlapping audio, in the present technical solution, the driver audio data set to be screened is also input into a pre-trained overlap detection model to split the overlapping audio and the driver audio data set, and finally obtain accurate driver audio data sets and passenger audio data sets, thereby solving the problem of inaccurate classification results due to factors such as voice overlap in the prior art.

[0095] Embodiment 2

[0096] The method also includes:

[0097] Obtain a driver audio sample dataset and an overlapping audio sample dataset;

[0098] Multiple driver audio sample datasets and an overlapping audio sample dataset are used as training datasets to train the overlap detection model.

[0099] It should be noted that obtaining the driver audio sample dataset includes:

[0100] The driver audio data collected when the driver registered and the historical driver audio dataset are obtained as the driver audio sample dataset.

[0101] In this embodiment, when a driver registers, the driver will be asked to read aloud preset texts such as "Hello", "Thank you", "You're welcome", etc. to collect initial driver audio data. However, these initial driver audio data cannot meet the requirements of the training model, so it is also necessary to obtain the historical driver audio data set generated by the driver's daily orders as sample data for the training model.

[0102] It should be noted that the historical driver audio dataset generated by drivers when accepting orders on a daily basis can better reflect the most realistic and accurate driver audio features.

[0103] The historical driver audio dataset is the driver audio dataset saved when the method is executed historically.

[0104] Therefore, the method also includes:

[0105] The driver audio data set is stored and marked with the corresponding date.

[0106] In this embodiment, the driver audio data set is stored in the vector database milvus on a daily basis to facilitate model calling and model updating.

[0107] It should be noted that the method also includes:

[0108] Standardize the driver audio sample dataset; including:

[0109] Segmenting the driver audio sample data in the driver audio sample data set;

[0110] Performing Mel-spectrogram conversion on the segmented driver audio sample data;

[0111] Calculate the mean and standard deviation of the driver audio sample data after the Mel-spectrogram conversion and perform normalization.

[0112] The driver audio sample data is segmented, including:

[0113] The first part of the audio data in the driver audio sample data set is divided based on the second audio duration, and the remaining part is divided based on the third audio duration;

[0114] The duration of the second audio is four times the duration of the third audio.

[0115] For example:

[0116] For a segment of driver audio sample data, the first segmentation is performed after an interval of 1280 milliseconds at the beginning, and then a new segmentation is performed each time a slide of 320 milliseconds is performed.

[0117] The driver audio sample data is segmented here mainly to limit the audio length and prevent video memory overflow, while ensuring processing consistency and reducing the computational burden.

[0118] By calculating the Mel-spectrogram and converting the driver audio sample data into decibel units, the essential characteristics of the audio can be captured more effectively. The use of decibel units helps standardize the dynamic range of the audio signal, making it more suitable for subsequent analysis.

[0119] Calculate the mean and standard deviation of the audio features and perform normalization. This step standardizes the feature scale between different audios and provides consistent and comparable input data for subsequent models.

[0120] It should be noted that training the overlap detection model includes:

[0121] Build an overlap detection model;

[0122] The training dataset is input into the overlap detection model, and it is iterated based on the contrastive learning loss function and the cross entropy loss function;

[0123] After the iteration is completed, the overlap detection model is evaluated, and the training is completed after the evaluation passes.

[0124] In this technical solution, the training data set of the overlap detection model includes multiple driver audio sample data sets and one overlap audio sample data set. The overlap audio sample data set is introduced to enable the overlap detection model to distinguish between overlap audio and driver audio. One overlap audio sample data set and multiple driver audio sample data sets are used as training data sets to prevent interference from some passenger audio and noise in the driver audio sample data set, thereby making the model more accurate.

[0125] The overlap detection model is trained to distinguish whether the input audio is the clear voice of a single speaker or a mixed voice containing multiple speakers.

[0126] Furthermore, the method also includes:

[0127] At every preset period, the overlap detection model is updated through the driver audio data set within the period.

[0128] Preferably, the preset period is set to 1 month. Since a new driver audio data set is generated every day when the technical solution is executed, when the new driver audio data set accumulates to a certain extent, the overlap detection model is updated by the new driver audio data set, thereby improving the accuracy of the overlap detection model.

[0129] Embodiment 3

[0130] A voiceprint clustering-based audio role recognition device for online car-hailing, referring to Figure 3 ,include:

[0131] Processor 21 and memory 22;

[0132] The processor 21 and the memory 22 are connected via a communication bus:

[0133] The processor 21 is used to call and execute the program stored in the memory 22;

[0134] The memory 22 is used to store programs, and the programs are used to execute at least one of the online car-hailing audio role recognition methods based on voiceprint clustering in the above embodiments.

[0135] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0136] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0137] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0138] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0140] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0141] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0142] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0143] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for audio role recognition of online car-hailing based on voiceprint clustering, characterized in that: include: Obtain the original audio data recorded by online ride-hailing vehicles during the day; Extracting human voice audio data from the original audio data; Segmenting the human voice audio data to obtain a plurality of human voice audio data segments; Clustering the human voice audio data segments to obtain a driver audio data set and a passenger audio data set to be screened; Inputting the driver audio data set to be screened into a pre-trained overlap detection model, and separating the overlapping audio and the driver audio data set from the driver audio data set to be screened by the overlap detection model; Clustering the human voice audio data segments to obtain a driver audio data set and a passenger audio data set to be screened includes: Determine the cosine similarity between two consecutive human voice audio data segments, and merge them if the cosine similarity between the two consecutive human voice audio data segments is higher than a first preset threshold; When the cosine similarity between any two consecutive human voice audio data segments is not higher than a first preset threshold, clustering the current audio data segments, and dividing the clusters longer than the first audio duration into a first cluster data set, and dividing the clusters not longer than the first audio duration into a second cluster data set; Take the average of all embedding vectors in each cluster dataset to get the central embedding vector; According to the central embedding vector of each cluster data set, the cosine similarity between the cluster data sets is determined; When the cosine similarity between the second type of clustering data set and any first type of clustering data set is higher than a second preset threshold, merge them into the first type of clustering data set with the highest cosine similarity; When the cosine similarity between the second type of clustered data set and any first type of clustered data set is not higher than a second preset threshold, determining it as a passenger audio data set; The non-passenger audio data set is determined as the driver audio data set to be screened.

2. The method according to claim 1, characterized in that: The method further comprises: Obtain a driver audio sample dataset and an overlapping audio sample dataset; The overlap detection model is trained by using a plurality of driver audio sample data sets and an overlap audio sample data set as training data sets.

3. The method according to claim 2, characterized in that Get the driver audio sample dataset, including: The driver audio data collected when the driver registers and the historical driver audio data set are obtained as the driver audio sample data set.

4. The method according to claim 2, characterized in that: The method further comprises: The driver audio sample data set is standardized; including: Segmenting the driver audio sample data in the driver audio sample data set; Performing Mel-spectrogram conversion on the segmented driver audio sample data; Calculate the mean and standard deviation of the driver audio sample data after the Mel-spectrogram conversion and perform normalization.

5. The method according to claim 2, characterized in that: Training the overlap detection model includes: Build an overlap detection model; Inputting the training data set into an overlap detection model, and iterating based on a contrastive learning loss function and a cross entropy loss function; After the iteration is completed, the overlap detection model is evaluated, and the training is completed after the evaluation passes.

6. The method according to claim 4, characterized in that Segmenting the human voice audio data includes: The first part of the human voice audio data is divided based on the second audio duration, and the remaining part is divided based on the third audio duration; The driver audio sample data is segmented, including: The first part of the audio data in the driver audio sample data set is divided based on the second audio duration, and the remaining part is divided based on the third audio duration; The duration of the second audio is four times the duration of the third audio.

7. The method according to claim 1, characterized in that The method further comprises: At every preset period, the overlap detection model is updated using the driver audio data set within the period.

8. The method according to claim 1, characterized in that The method further comprises: The driver audio data set is stored and marked with the corresponding date.

9. An audio role recognition device for online car-hailing based on voiceprint clustering, characterized in that: include: Processor and memory; The processor and the memory are connected via a communication bus: Wherein, the processor is used to call and execute the program stored in the memory; The memory is used to store a program, and the program is at least used to execute the method for audio role recognition of online car-hailing based on voiceprint clustering as described in any one of claims 1-8.

Images