Audio processing method and device, equipment, storage medium and program product

Abstract

The embodiment of the invention provides an audio processing method and device, equipment, a storage medium and a program product. The method comprises the following steps: constructing structured data associated with an audio data stream; based on the structured data, multiple voiceprint clustering results are generated, and the multiple voiceprint clustering results correspond to different numbers of voiceprint labels; and based on evaluation information corresponding to the various voiceprint clustering results, determining voiceprint information associated with the audio data stream, the voiceprint information indicating the voiceprint feature representation and the voiceprint tag of at least one speaker in the audio data stream. In this way, the accuracy of identifying the voiceprint label of the speaker in the audio data stream can be improved.

Description

TECHNICAL FIELD

[0001] Example embodiments of the present disclosure generally relate to the field of computers, and in particular, to a method, an apparatus, an equipment, a computer readable storage medium and a computer program product for audio processing. BACKGROUND

[0002] Speaker diarization is an audio processing task aiming to automatically segment an audio stream containing multiple speakers into several segments and label each segment according to the identity of the speaker, so as to distinguish the speech of different speakers. However, the existing processing speaker diarization task usually adopts a simple clustering method to cluster the voiceprint features, resulting in low accuracy. SUMMARY

[0003] In a first aspect of the present disclosure, a method for audio processing is provided. The method comprises: constructing structured data associated with an audio data stream; generating a plurality of voiceprint clustering results based on the structured data, the plurality of voiceprint clustering results corresponding to different numbers of voiceprint labels; and determining voiceprint information associated with the audio data stream based on evaluation information corresponding to the plurality of voiceprint clustering results, the voiceprint information indicating a voiceprint feature representation and a voiceprint label of at least one speaker in the audio data stream.

[0004] In a second aspect of the present disclosure, an apparatus for audio processing is provided. The apparatus comprises: a construction module configured to construct structured data associated with an audio data stream; a generation module configured to generate a plurality of voiceprint clustering results based on the structured data, the plurality of voiceprint clustering results corresponding to different numbers of voiceprint labels; and a determination module configured to determine voiceprint information associated with the audio data stream based on evaluation information corresponding to the plurality of voiceprint clustering results, the voiceprint information indicating a voiceprint feature representation and a voiceprint label of at least one speaker in the audio data stream.

[0005] In a third aspect of the present disclosure, an electronic device is provided. The device comprises at least one processor; and at least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor. The instructions, when executed by the at least one processor, cause the device to perform the method of the first aspect.

[0006] In a fourth aspect of the present disclosure, a computer readable storage medium is provided. The computer readable storage medium has stored thereon computer executable instructions executable by a processor to implement the method of the first aspect.

[0007] In a fifth aspect of the disclosure, a computer program product is provided. The computer program product is tangibly stored in a computer storage medium and includes computer executable instructions that, when executed by a device, cause the device to perform the method of the first aspect.

[0008] It is to be understood that the details set forth herein are by way of example and not intended to limit the scope of the embodiments of the present disclosure, but rather to provide detailed descriptions of various embodiments of the disclosure. It should be further understood that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting. BRIEF DESCRIPTION OF DRAWINGS

[0009] The above and other features, aspects, and advantages of various embodiments of the present disclosure will become more apparent from the following detailed description, taken in conjunction with the accompanying drawings, in which like reference numbers represent like elements throughout. In the drawings:

[0010] FIG. 1 A schematic diagram illustrating an example environment in which embodiments of the present disclosure can be implemented is shown.

[0011] FIG. 2 A flow diagram illustrating an example process of audio processing according to some embodiments of the present disclosure is shown;

[0012] FIG. 3 A flow diagram illustrating an example of audio recognition according to some embodiments of the present disclosure is shown;

[0013] FIG. 4 A schematic diagram illustrating a voiceprint matrix according to some embodiments of the present disclosure is shown;

[0014] FIG. 5A And FIG. 5B A schematic diagram illustrating an embodiment of audio recognition according to some embodiments of the present disclosure is shown;

[0015] FIG. 6 A schematic block diagram illustrating an example apparatus for audio processing according to some embodiments of the present disclosure is shown; and

[0016] FIG. 7 A block diagram of an electronic device capable of implementing various embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0017] Embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings. While several embodiments of the present disclosure are described, it should be understood that the disclosure can be embodied in many other forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and fully convey the scope of the disclosure to those skilled in the art.

[0018] It is noted that the headings provided herein are not limitations of various embodiments provided herein. Various embodiments are described throughout this document and any type of embodiment can be included under any section. Moreover, embodiments described in any section can be combined with any other embodiment described in the same section and / or a different section in any manner.

[0019] In the description of embodiments of the disclosure, the term "includes" and its derivatives, such as "including" should be understood in an open way, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". The following can also include other explicit and implicit definitions. The terms "first", "second", etc. can refer to different or same objects. The following can also include other explicit and implicit definitions.

[0020] Data of users, acquisition and / or use of data, etc. can be involved in embodiments of the disclosure. These aspects all comply with corresponding laws and regulations and relevant provisions. In embodiments of the disclosure, all data collection, acquisition, processing, processing, forwarding, use, etc. are performed on the premise that users are aware of and confirm. Accordingly, when implementing embodiments of the disclosure, the type of data or information that can be involved, the use range, the use scenario, etc. should be notified to the user and the authorization of the user should be obtained through appropriate means according to relevant laws and regulations. The specific notification and / or authorization manner can vary according to actual situations and application scenarios, and the scope of the disclosure is not limited in this aspect.

[0021] In the specification and embodiments, if personal information processing is involved, it will be processed on the premise of legality (for example, obtaining the consent of the subject of personal information, or being necessary for the performance of a contract, etc.), and only within the prescribed or agreed range. Users refuse to process personal information other than the necessary information required for basic functions, which will not affect the user's use of basic functions.

[0022] As briefly mentioned above, when performing a speaker diary task according to the prior art, a high delay is generated, such as a long time is needed to detect the label corresponding to the new speaker after the new speaker speaks. Moreover, due to the use of a relatively simple clustering method in the prior art, the accuracy of the detection result is affected, such as a speaker is clustered into two labels, or multiple speakers are clustered into the same label.

[0023] To this end, embodiments of the present disclosure propose a scheme of audio processing. According to the scheme, structured data associated with an audio data stream is constructed; based on the structured data, a plurality of speaker diarization results corresponding to different numbers of speaker labels are generated; and based on evaluation information corresponding to the plurality of speaker diarization results, speaker information associated with the audio data stream is determined, the speaker information indicating a speaker characteristic representation and a speaker label of at least one speaker in the audio data stream.

[0024] In this way, by generating a plurality of speaker diarization results corresponding to different numbers of speaker labels, and by evaluating the plurality of speaker diarization results through evaluation information, a more accurate determination of the number of speakers in the audio data stream can be achieved, thereby performing accurate clustering and obtaining more accurate speaker information associated with the audio data stream.

[0025] Example embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0026] Example Environment

[0027] FIG. 1 A schematic diagram illustrating an example environment in which embodiments of the present disclosure can be implemented is shown. FIG. 1 A schematic diagram illustrating an example environment 100 in which embodiments of the present disclosure can be implemented is shown. As shown, the example environment 100 can include an electronic device 110 and a model 120. The electronic device 110 can implement the above-mentioned scheme of audio processing through the model 120. The model 120 can be deployed on the electronic device 110, and can also be deployed on other devices, which will not be described herein. FIG. 1

[0028] The electronic device 110 can be any type of mobile terminal, fixed terminal, or portable terminal including a mobile handset, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a media player, a multimedia tablet, a palmtop computer, a portable gaming terminal, a VR / AR device, a Personal Communication System (PCS) device, a personal navigation device, a Personal Digital Assistant (PDA), an audio / video player, a digital camera / camcorder, a positioning device, a television receiver, a radio broadcast receiver, an electronic book device, a game device, or any combination thereof, including an accessory or peripheral device thereof or any combination thereof. In some embodiments, the electronic device 110 can also support any type of interface to a target user (such as "wearable" circuitry, etc.).

[0029] ​The electronic device 110 can also be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, a cloud server providing cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content distribution network, and basic cloud computing services such as big data and artificial intelligence platform, etc. The electronic device 110 may, for example, include a computing system / server, such as a mainframe, an edge computing node, a computing device in a cloud environment, etc.

[0030] It should be understood that the structure and function of the various elements in the environment 100 are described for illustrative purposes only, without implying any limitation on the scope of the present disclosure.

[0031] Example Process

[0032] FIG. 2 A flowchart of an example process of audio processing according to some embodiments of the present disclosure is shown. The process 200 can be implemented at the electronic device 110.

[0033] As shown in block 210, the electronic device 110 can construct structured data associated with the audio data stream. FIG. 2 For example, an audio data stream can be obtained, in which the speech of at least one speaker is included. Referring to step 301 in

[0034] For the convenience of identifying operations on the audio data stream, the audio data stream can be segmented into multiple segments in real time, such as cutting the audio in sentence units. In order to accurately identify the speech intervals of different speakers from the audio data stream and assign a unique label to each speaker, the structured data associated with the audio data stream can be constructed according to step 302 in FIG. 3 FIG. 3 For example, a voiceprint matrix including a first matrix 401, a second matrix 402, a third matrix 403, and a fourth matrix 404 can be constructed, and then the structured data can be obtained by performing feature dimension reduction operation on the voiceprint matrix. Then, referring to step 303 in FIG. 4 FIG. 3

[0035] ​​​In some embodiments, the electronic device 110 can obtain a voiceprint matrix corresponding to the audio data stream, the voiceprint matrix including a first number of dimensions. The voiceprint matrix is then mapped to structured data, the structured data including a second number of dimensions, and the second number is less than the first number. For example, since the voiceprint matrix corresponding to the audio data stream usually includes multiple dimensions, the high-dimensional voiceprint matrix can be nonlinearly mapped by a data dimension reduction technique to obtain low-dimensional structured data, thereby capturing the local structure and complex nonlinear relationship of the voiceprint matrix.

[0036] In some embodiments, the reference FIG. 4 The voiceprint matrix can include multiple parts, the first matrix 401, the second matrix 402, the third matrix 403, and the fourth matrix 404. The electronic device 110 can determine at least one preset voiceprint feature representation. Then add first perturbation information to the at least one preset voiceprint feature representation to construct the first matrix.

[0037] For example, at least one preset voiceprint feature representation can be determined first, which can be an artificially constructed voiceprint feature representation. Then the first matrix dec can be constructed by adding first perturbation information eps to the artificially constructed voiceprint feature representation, the number of which can be dn. And the first matrix can include 2 sub-matrices, representing the voiceprint feature representations of two virtual speakers, and each sub-matrix includes 30 voiceprint feature vectors. The i-th voiceprint in the first sub-matrix in the first matrix can be generated by the following formula:

[0038] dec1(i) = 0.1 ed + eps ed * randn(0, 1)

[0039] The i-th voiceprint in the second sub-matrix in the first matrix can be generated by the following formula:

[0040] dec2(i) = -0.1 ed + eps ed * randn(0, 1)

[0041] By introducing the first matrix, the clustering accuracy in the case of insufficient data volume can be improved, thereby solving the problem of high delay when a new speaker appears in the audio data stream.

[0042] In some embodiments, the voiceprint matrix can also include a second matrix, and the electronic device 110 can obtain a historical voiceprint set, the historical voiceprint set including a mapping between a voiceprint label of at least one historical speaker and at least one voiceprint feature representation. Then construct the second matrix based on the historical voiceprint set.

[0043] For example, a historical voiceprint set can be acquired first, in which the voiceprint labels of at least one historical speaker and the voiceprint feature representations corresponding to the voiceprint labels are stored. Then the second matrix can be constructed by splicing the plurality of voiceprint feature representations stored in the historical voiceprint set. In this way, the newly added speaker in the audio data stream can be quickly detected, and the accuracy of the overall audio recognition can be improved.

[0044] In some embodiments, the historical voiceprint set can be updated as the audio data stream is updated. The electronic device 110 can determine the voiceprint label and the voiceprint feature representation corresponding to the at least one newly added speaker in response to detecting the at least one newly added speaker in the updated audio data stream. Then the historical voiceprint set is updated based on the voiceprint label and the voiceprint feature representation corresponding to the at least one newly added speaker.

[0045] For example, if speaker A and speaker B are speaking in the audio data stream, and later speaker C also joins in the speaking, the voiceprint label and the corresponding voiceprint feature representation of speaker C can be stored in the historical voiceprint set. In addition, since the historical voiceprint set is updated as the audio data stream is updated, the information stored in the historical voiceprint set will become more and more, and an upper limit can be set for the information stored in the historical voiceprint set, for example, for each voiceprint label, the maximum number of voiceprint feature representations stored is 30.

[0046] In some embodiments, the voiceprint matrix can further include a third matrix, and the electronic device 110 can determine a plurality of continuous voiceprint feature representations from the audio data stream based on a preset window length t sen. Then the third matrix t sec is constructed based on the plurality of continuous voiceprint feature representations. For example, a sliding voiceprint window with a length of 20 can be determined, and 20 continuous voiceprint feature representations can be acquired from the audio data stream by applying the sliding voiceprint window, and the continuous voiceprint feature representations include the voiceprint feature representation ce corresponding to the current time. Then the third matrix is constructed according to the 20 continuous voiceprint feature representations. In this way, the time continuity feature of the voiceprint can be provided.

[0047] In some embodiments, the voiceprint matrix can further include a fourth matrix sec, and the electronic device 110 can add second perturbation information to the voiceprint feature representation corresponding to the current segment of the audio data stream to construct the fourth matrix. For example, in order to facilitate the recognition operation of the audio, the audio data stream can be divided into a plurality of segments in real time, for example, the audio can be divided in units of sentences. Then the second perturbation information can be added to the voiceprint feature representation corresponding to the latest segment to construct the fourth matrix, and the number of the fourth matrix can be sn, for example, 20. The i th column in the fourth matrix can be generated by the following formula:

[0048] sec(i) = ce + epsed *randn(0,1)

[0049] By this method, the problem of clustering accuracy when adding a new speaker can be solved.

[0050] In some embodiments, after obtaining the first matrix, the second matrix, the third matrix and the fourth matrix, these matrices can be spliced to obtain the final voiceprint matrix dataset = [dec1, dec2, hec, tsec, sec]. In addition, the voiceprint matrix in the embodiments of the present disclosure is not limited to the construction method described above, and one or more matrices can be selected from the first matrix, the second matrix, the third matrix and the fourth matrix to construct the voiceprint matrix according to actual conditions.

[0051] In block 220, the electronic device 110 can generate a plurality of voiceprint clustering results based on the structured data.

[0052] In some embodiments, when clustering the voiceprint feature representations of the audio data stream, the number of clusters needs to be specified in advance. For example, if there are 7 speakers speaking in the audio data stream, the number of clusters needs to be specified as 7. If the number is specified incorrectly, such as being specified as 9, the clustering accuracy will be greatly reduced. Therefore, it is very important to determine the appropriate number of clusters. For example, a plurality of numbers can be determined in advance, and a plurality of voiceprint clustering results corresponding to the plurality of numbers can be generated based on the structured data, and the number of speakers in the audio data stream can be finally determined through different voiceprint clustering results.

[0053] In some embodiments, the electronic device 110 can determine a plurality of clustering parameters, the clustering parameters indicating the number of speakers in the audio data stream. Then, the structured data is clustered based on the plurality of clustering parameters respectively, to obtain a plurality of voiceprint label sets corresponding to the plurality of clustering numbers. Next, a plurality of voiceprint clustering results are constructed based on the plurality of voiceprint label sets.

[0054] For example, a plurality of clustering parameters such as 6, 7 and 8 can be determined first. Then, the structured data is clustered according to these clustering parameters respectively, to obtain a voiceprint label set corresponding to the clustering parameter 6, which includes 6 voiceprint labels and a plurality of voiceprint feature representations corresponding to the 6 voiceprint labels; a voiceprint label set corresponding to the clustering parameter 7, which includes 7 voiceprint labels and a plurality of voiceprint feature representations corresponding to the 7 voiceprint labels; and a voiceprint label set corresponding to the clustering parameter 8, which includes 8 voiceprint labels and a plurality of voiceprint feature representations corresponding to the 8 voiceprint labels. Then, a plurality of voiceprint clustering results can be constructed according to these voiceprint label sets.

[0055] In some embodiments, the clustering parameters can include at least one of: a first parameter indicating at least one preset number of speakers in the audio data stream. A second parameter indicating a number of speakers in the historical audio data stream.

[0056] For example, if the audio data stream is from a conference room, the maximum number of people lc+1 that the conference room can accommodate, such as 3, and the minimum number of people lc-1, such as 1, can be obtained, as well as the number of people participating in the last meeting of the conference room, that is, the number of speakers in the historical audio data stream, such as 3. Then the clustering parameters are specified as all values between the maximum number and the minimum number, such as 1, 2 and 3, and if the number of speakers in the historical audio data stream is not included in these values, the clustering parameters can also be specified as the number of speakers in the historical audio data stream. Then the voiceprint clustering results corresponding to these clustering parameters are obtained.

[0057] In block 230, the electronic device 110 can determine the voiceprint information associated with the audio data stream based on the evaluation information corresponding to the plurality of voiceprint clustering results.

[0058] For example, after obtaining the plurality of voiceprint clustering results, the evaluation information corresponding to each voiceprint clustering result can be obtained, and the number of speakers in the current audio data stream can be determined according to the evaluation information, and finally the voiceprint information associated with the audio data stream is determined, such as the voiceprint information can include n speakers in the current audio data stream, the voiceprint label corresponding to each speaker, and the voiceprint feature representation corresponding to each voiceprint label.

[0059] In some embodiments, the electronic device 110 can determine a plurality of evaluation information corresponding to the plurality of voiceprint clustering results. Then based on the plurality of evaluation information, a target number of speakers in the audio data stream is determined. Next, based on the voiceprint clustering result corresponding to the target number, the voiceprint information associated with the audio data stream is determined. For example, the evaluation information corresponding to each voiceprint clustering result, such as the silhouette coefficient, can be obtained. Then according to the size of the numerical value of these evaluation information, the target number of speakers in the audio data stream is determined. Next, according to the voiceprint clustering result corresponding to the target number, the final voiceprint information is determined. The silhouette coefficient is an index for evaluating the clustering effect, which combines the cohesion and separation of clustering.

[0060] In some embodiments, the electronic device 110 can determine the similarity between the feature representations in the same cluster and the difference between the feature representations in different clusters in the voiceprint clustering result based on the voiceprint feature representations corresponding to the plurality of voiceprint labels in the voiceprint clustering result. Then based on the similarity and the difference, a plurality of evaluation information corresponding to the plurality of voiceprint clustering results is determined.

[0061] For example, since each voiceprint clustering result includes voiceprint feature representations corresponding to multiple voiceprint labels, the centroid of the voiceprint feature representation cluster corresponding to each voiceprint label can be obtained, and the distance between each voiceprint feature representation in the voiceprint feature representation cluster and the centroid can be calculated. According to these distances, the similarity between the feature representations in the same cluster can be determined. Similarly, the average distance between the voiceprint feature representation clusters can also be calculated to determine the difference between the feature representations in different clusters. Then, according to the similarity and the difference, the plurality of contour coefficients corresponding to the plurality of voiceprint clustering results can be determined.

[0062] In some embodiments, the electronic device 110 can determine, from the plurality of evaluation information, a first evaluation information and a second evaluation information obtained based on the number of speakers in the historical audio data stream, the first evaluation information being superior to the second evaluation information. Then, the first evaluation information is updated based on a preset update coefficient. Next, the target number of speakers in the audio data stream is determined based on the first evaluation information and the second evaluation information.

[0063] For example, after obtaining the plurality of evaluation information, the second evaluation information obtained based on the number of speakers in the historical audio data stream and the first evaluation information with the largest value in the plurality of evaluation information can be determined. Then, a penalty coefficient can be introduced into the first evaluation information, and the first evaluation information is updated according to the penalty coefficient. Then, the value of the updated first evaluation information and the value of the second evaluation information are compared, and the number of clustering parameters corresponding to the evaluation information with the larger value is determined as the target number of speakers in the audio data stream.

[0064] In some embodiments, the electronic device 110 can determine the target number of speakers in the audio data stream based on the clustering parameters corresponding to the first evaluation information in response to the first evaluation information being superior to the second evaluation information. In response to the first evaluation information not being superior to the second evaluation information, the number of speakers in the historical audio data stream is determined as the target number of speakers in the audio data stream.

[0065] For example, if the value of the first evaluation information after introducing the penalty coefficient is greater than the value of the second evaluation information, the clustering parameters corresponding to the first evaluation information can be determined as the target number of speakers in the audio data stream. If the value of the first evaluation information after introducing the penalty coefficient is not greater than the value of the second evaluation information, the number of speakers in the historical audio data stream can be determined as the target number of speakers in the audio data stream.

[0066] In some embodiments, after determining the target number of speakers in the audio data stream, the voiceprint clustering result corresponding to the target number of speakers can be obtained, and the voiceprint information associated with the audio data stream can be determined according to the voiceprint clustering result. After clustering, the voiceprint labels need to be reorganized to ensure that the voiceprint label numbers start from 0.

[0067] In some embodiments, to further improve clustering accuracy, the voiceprint quality can be evaluated again after clustering is completed to determine whether to add the current voiceprint to the historical voiceprint set and whether to update the speaker label corresponding to the current voiceprint. The electronic device 110 can obtain the historical voiceprint set and the voiceprint feature representation and voiceprint label corresponding to at least one current speaker in the current segment in the audio data stream. Then, based on the voiceprint feature representation and voiceprint label corresponding to the current speaker, a target label matched by the current speaker is determined. Subsequently, based on the historical voiceprint set and the voiceprint feature representation and voiceprint label corresponding to the current speaker, a plurality of correlations between the current speaker and the historical speakers are determined. Next, based on the plurality of correlations, the historical voiceprint set and the target label are updated.

[0068] For example, referring to step 501 in FIG. 5A , the historical voiceprint set and the voiceprint feature representation and voiceprint label corresponding to at least one current speaker in the current segment in the audio data stream can be obtained. Then, based on the voiceprint feature representation and voiceprint label corresponding to the current speaker, a target label matched by the current speaker is determined from at least one voiceprint label in the historical voiceprint set. Subsequently, a plurality of correlations between the current speaker and the historical speakers are generated, and next, based on the plurality of correlations, it is determined whether to update the historical voiceprint set and the target label.

[0069] In some embodiments, the electronic device 110 can determine at least one target feature representation corresponding to at least one historical speaker based on the historical voiceprint set. Then, based on the voiceprint feature representation corresponding to at least one speaker in the current segment and the at least one target feature representation, a correlation between the voiceprint feature representation of the at least one current speaker and the voiceprint feature representation of the at least one historical speaker is determined.

[0070] For example, referring to step 502 in FIG. 5A , since each voiceprint label in the historical voiceprint set corresponds to a plurality of voiceprint feature representations, the plurality of voiceprint feature representations corresponding to each voiceprint label can be regarded as a cluster, and the centroid in the cluster can be taken as the target feature representation centroid key . The calculation method of the centroid is to add the plurality of voiceprint feature representations and take the average.

[0071] Then, referring to step 503 in FIG. 5AIn step 503, a correlation between the voiceprint feature representation corresponding to the at least one current speaker in the current segment and the target feature representation is calculated, and the correlation represents a degree of correlation between the current speaker and the historical speaker. A correlation threshold thres can be set, and if the number of voiceprint feature representations corresponding to the voiceprint labels in the historical voiceprint set is less than 3, the correlation threshold can be set to 0.5, otherwise, the correlation threshold is set to 0.6. The correlation between the voiceprint feature representation ce corresponding to the current speaker and the target feature representation can be calculated by the following formula:

[0072]

[0073] In some embodiments, the electronic device 110 can update the historical voiceprint set and the target label based on the plurality of correlations between the current speaker and the historical speakers, in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label not satisfying the first preset condition.

[0074] For example, the current speaker is previously identified to correspond to the target label, and the target label corresponds to the historical speaker 1. The plurality of correlations between the current speaker and the historical speakers 1, 2, 3 and 4 in the historical voiceprint set are obtained. Referring to FIG. 5A In step 504, if the correlation between the current speaker and the historical speaker 1 is not the largest in the plurality of correlations, it indicates that the matching relationship between the current speaker and the historical speaker corresponding to the target label does not satisfy the first preset condition, that is, the label of the current speaker matching the historical speaker 1 is not appropriate, and the previous matching may be incorrect and needs to be adjusted.

[0075] In some embodiments, the electronic device 110 can update the target label in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label not satisfying the first preset condition, and the correlation between the current speaker and the historical speaker corresponding to the target label does not exceed the first threshold. Then, the historical voiceprint set is updated based on the target label and the voiceprint feature representation of the current speaker.

[0076] For example, referring to FIG. 5A In steps 507 to 508, if the correlation between the current speaker and the at least one historical speaker is less than 0.3, it indicates that it is unreasonable to match the current speaker with the labels of these historical speakers, and the current speaker can be a new speaker that is not stored in the historical voiceprint set. Therefore, a new label can be assigned to the current speaker, and the historical voiceprint set is updated according to the new label and the voiceprint feature representation corresponding to the current speaker, that is, the new speaker is stored in the historical voiceprint set.

[0077] In some embodiments, the electronic device 110 can update the target label based on the voiceprint label in the historical voiceprint set in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label not satisfying the first preset condition, and the relevance between the current speaker and the historical speaker corresponding to the target label exceeding the second threshold. Then, the historical voiceprint set is updated based on the target label and the voiceprint feature representation of the current speaker.

[0078] For example, referring to steps 509-510 in FIG. 5A For example, referring to steps 509-510 in FIG. 5A For example, referring to steps 509-510 in

[0079] In some embodiments, since the number of voiceprint feature representations stored in the historical voiceprint set is limited by an upper limit, it is necessary to determine whether to continue data storage according to the storage condition in the historical voiceprint set. The electronic device 110 can determine the voiceprint density corresponding to the target label in response to the number of voiceprint feature representations corresponding to the target label in the historical voiceprint set exceeding a preset storage number. Then, the edge feature representation is determined from the voiceprint feature representations corresponding to the target label based on the voiceprint density. After that, the historical voiceprint set is updated based on the edge feature representation and the voiceprint feature representation of the current speaker.

[0080] For example, referring to steps 513 and 515 in FIG. 5B For example, referring to steps 513 and 515 in FIG. 5B For example, referring to steps 513 and 515 in

[0081] For example, referring to steps 513 and 515 in FIG. 5BAt step 514 in FIG. 5, if the target label is the label corresponding to historical speaker 2, and the number of voiceprint feature representations corresponding to the label of historical speaker 2 does not exceed the upper limit 30, the voiceprint feature representation of the current speaker can be directly stored into the historical voiceprint set.

[0082] In some embodiments, the electronic device 110 can update the historical voiceprint set based on the plurality of correlations between the current speaker and the historical speaker corresponding to the target label in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label satisfying the first preset condition. For example, referring to steps 505 and 506 in FIG. 5, if it is identified that the current speaker corresponds to the target label, and the target label corresponds to historical speaker 1, and the matching relationship between the current speaker and historical speaker 1 satisfies the first preset condition, it indicates that the target label is correct, and the target label does not need to be updated. If the correlation between the current speaker and historical speaker 1 exceeds the second threshold, the voiceprint feature representation of the current speaker can be stored into the historical voiceprint set. FIG. 5A

[0083] In some embodiments, the electronic device 110 can update the historical voiceprint set based on the plurality of correlations between the current speaker and the historical speaker corresponding to the target label in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label satisfying the first preset condition. For example, referring to steps 505 and 506 in FIG. 5, if it is identified that the current speaker corresponds to the target label, and the target label corresponds to historical speaker 1, and the matching relationship between the current speaker and historical speaker 1 satisfies the first preset condition, it indicates that the target label is correct, and the target label does not need to be updated. If the correlation between the current speaker and historical speaker 1 exceeds the second threshold, the voiceprint feature representation of the current speaker can be stored into the historical voiceprint set. FIG. 5A

[0084] For another example, if the number of voiceprint feature representations corresponding to the label of historical speaker 1 exceeds the upper limit 30, the voiceprint density is determined according to the voiceprint feature representations of historical speaker 1. The most marginal voiceprint feature representation, i.e., the marginal feature representation, is determined from the plurality of voiceprint feature representations according to the voiceprint density. Then the distance between the marginal feature representation and the centroid, and the distance between the voiceprint feature representation of the current speaker and the centroid are calculated respectively. If the marginal feature representation is closer to the centroid, the historical voiceprint set does not need to be updated. If the voiceprint feature representation of the current speaker is closer to the centroid, the voiceprint feature representation of the current speaker is used to replace the marginal feature representation in the historical voiceprint set. In this way, the information stored in the historical voiceprint set can be more accurate.

[0085] For another example, if the target label is the label corresponding to historical speaker 1, and the number of voiceprint feature representations corresponding to the label of historical speaker 1 does not exceed the upper limit 30, the voiceprint feature representation of the current speaker can be directly stored into the historical voiceprint set.

[0086] ​​In this way, the structured data corresponding to the audio data stream can be clustered according to multiple clustering parameters respectively, and multiple voiceprint clustering results can be obtained, so as to accurately determine the number of speakers in the audio data stream at the current moment. Moreover, the target label of the current speaker and the historical voiceprint set can be updated after clustering, so as to store more accurate data in the historical voiceprint set, match a more suitable label for the current speaker, and improve the accuracy of the overall audio recognition.

[0087] Example Devices and Apparatus

[0088] Embodiments of the present disclosure also provide a corresponding apparatus for implementing the above method or process. FIG. 6 A schematic structural block diagram of an apparatus 600 for audio processing according to certain embodiments of the present disclosure is shown. The apparatus 600 can be implemented as or included in the electronic device 110. Various modules / components in the apparatus 600 can be implemented by hardware, software, firmware, or any combination thereof.

[0089] As shown in FIG. 6 The apparatus 600 includes a construction module 610 configured to construct structured data associated with an audio data stream; a generation module 620 configured to generate multiple voiceprint clustering results based on the structured data, the multiple voiceprint clustering results corresponding to different numbers of voiceprint labels; and a determination module 630 configured to determine voiceprint information associated with the audio data stream based on evaluation information corresponding to the multiple voiceprint clustering results, the voiceprint information indicating a voiceprint feature representation and a voiceprint label of at least one speaker in the audio data stream.

[0090] In some embodiments, the construction module 610 is configured to obtain a voiceprint matrix corresponding to the audio data stream, the voiceprint matrix including a first number of dimensions; and map the voiceprint matrix into the structured data, the structured data including a second number of dimensions, and the second number being less than the first number.

[0091] In some embodiments, the construction module 610 is configured to determine at least one preset voiceprint feature representation; and add first perturbation information to the at least one preset voiceprint feature representation to construct a first matrix.

[0092] In some embodiments, the construction module 610 is configured to obtain a historical voiceprint set, the historical voiceprint set including a mapping between a voiceprint label and a voiceprint feature representation of at least one historical speaker; and construct a second matrix based on the historical voiceprint set.

[0093] In some embodiments, the constructing module 610 is configured to determine a plurality of continuous voiceprint feature representations from the audio data stream based on a preset window length, the plurality of continuous voiceprint feature representations including a voiceprint feature representation corresponding to a current time; and construct a third matrix based on the plurality of continuous voiceprint feature representations.

[0094] In some embodiments, the constructing module 610 is configured to add second perturbation information to the voiceprint feature representation corresponding to the current segment of the audio data stream to construct a fourth matrix.

[0095] In some embodiments, the generating module 620 is configured to determine a plurality of clustering parameters, the clustering parameters indicating a number of speakers in the audio data stream; cluster the structured data based on the plurality of clustering parameters respectively to obtain a plurality of voiceprint label sets corresponding to a plurality of clustering numbers; and construct a plurality of voiceprint clustering results based on the plurality of voiceprint label sets.

[0096] In some embodiments, the clustering parameters include at least one of: a first parameter indicating at least one preset number of speakers in the audio data stream; and a second parameter indicating a number of speakers in a historical audio data stream.

[0097] In some embodiments, the determining module 630 is configured to determine a plurality of evaluation information corresponding to the plurality of voiceprint clustering results; determine a target number of speakers in the audio data stream based on the plurality of evaluation information; and determine voiceprint information associated with the audio data stream based on a voiceprint clustering result corresponding to the target number of speakers.

[0098] In some embodiments, the determining module 630 is configured to determine, from the plurality of evaluation information, a first evaluation information and a second evaluation information obtained based on a number of speakers in a historical audio data stream, the first evaluation information being superior to the second evaluation information; update the first evaluation information based on a preset update coefficient; and determine the target number of speakers in the audio data stream based on the first evaluation information and the second evaluation information.

[0099] In some embodiments, the determining module 630 is configured to, in response to the first evaluation information being superior to the second evaluation information, determine the target number of speakers in the audio data stream based on a clustering parameter corresponding to the first evaluation information; and in response to the first evaluation information not being superior to the second evaluation information, determine the number of speakers in the historical audio data stream as the target number of speakers in the audio data stream.

[0100] In some embodiments, the apparatus 600 is configured to obtain a historical voiceprint set, and a voiceprint feature representation and a voiceprint label corresponding to at least one current speaker in a current segment in the audio data stream; determine a target label matched by the current speaker based on the voiceprint feature representation and the voiceprint label corresponding to the current speaker; determine a plurality of correlations between the current speaker and historical speakers based on the historical voiceprint set, and the voiceprint feature representation and the voiceprint label corresponding to the current speaker; and update the historical voiceprint set and the target label based on the plurality of correlations.

[0101] In some embodiments, the apparatus 600 is configured to, in response to a matching relationship between the current speaker and a historical speaker corresponding to the target label not satisfying a first preset condition, update the historical voiceprint set and the target label based on the plurality of correlations between the current speaker and the historical speaker; and in response to the matching relationship between the current speaker and the historical speaker corresponding to the target label satisfying the first preset condition, update the historical voiceprint set based on the plurality of correlations between the current speaker and the historical speaker.

[0102] In some embodiments, the apparatus 600 is configured to, in response to a matching relationship between the current speaker and a historical speaker corresponding to the target label not satisfying a first preset condition, and a correlation between the current speaker and the historical speaker corresponding to the target label not exceeding a first threshold, update the target label; and update the historical voiceprint set based on the target label and a voiceprint feature representation of the current speaker.

[0103] In some embodiments, the apparatus 600 is configured to, in response to a matching relationship between the current speaker and a historical speaker corresponding to the target label not satisfying a first preset condition, and a correlation between the current speaker and the historical speaker corresponding to the target label exceeding a second threshold, update the target label based on a voiceprint label in the historical voiceprint set; and update the historical voiceprint set based on the target label and a voiceprint feature representation of the current speaker.

[0104] In some embodiments, the apparatus 600 is configured to, in response to a number of voiceprint feature representations corresponding to the target label in the historical voiceprint set exceeding a preset storage number, determine a voiceprint density corresponding to the target label; determine an edge feature representation from the voiceprint feature representations corresponding to the target label based on the voiceprint density; and update the historical voiceprint set based on the edge feature representation and a voiceprint feature representation of the current speaker.

[0105] The units included in the apparatus 600 can be implemented utilizing various means, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units can be implemented using software and / or firmware, e.g., machine-executable instructions stored on a storage medium. In addition to or alternatively, some or all of the units in the apparatus 600 can be implemented at least partially by one or more hardware logic components. As an example and not by way of limitation, example types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Application-specific Integrated Circuits (ASICs), Application-specific Standard Products (ASSPs), System-on-a-Chip (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0106] FIG. 7 A block diagram of an electronic device 700 in which one or more embodiments of the disclosure can be implemented is shown. It should be understood that FIG. 7 The electronic device 700 shown is merely exemplary and should not be construed as limiting the scope of the embodiments described herein. FIG. 7 The electronic device 700 shown can be used to implement FIG. 1 the electronic device 110.

[0107] As FIG. 7 shown, the electronic device 700 is in the form of a general electronic device. Components of the electronic device 700 can include, but are not limited to, at least one processor 710 or processing unit, a memory 720, a storage device 730, one or more communication units 740, one or more input devices 750, and one or more output devices 760. The processor 710 can be a real or virtual processor and capable of performing various processes according to programs stored in the memory 720. In a multi-processor system, multiple processors perform computer-executable instructions in parallel to improve parallel processing capabilities of the electronic device 700.

[0108] The electronic device 700 typically includes a number of computer storage media. Such media can be any available media that is accessible by the electronic device 700 and includes both volatile and non-volatile media, removable and non-removable media. The memory 720 can be a volatile memory (e.g., registers, cache, random access memory (RAM)), a non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. The storage device 730 can be a removable or non-removable medium and can include machine-readable media, such as a flash drive, a disk drive, or any other medium that can be used to store information and / or data and that can be accessed by the electronic device 700.

[0109] Electronic device 700 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not explicitly stated... FIG. 7 As shown, disk drives for reading from or writing to removable, non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable, non-volatile optical disks can be provided. In these cases, each drive can be connected to a bus (not shown) via one or more data media interfaces. Memory 720 may include computer program product 725 having one or more program modules configured to perform various methods or actions of various embodiments of this disclosure.

[0110] The communication unit 740 enables communication with other electronic devices via a communication medium. Additionally, the functionality of the components of the electronic device 700 can be implemented using a single computing cluster or multiple computing machines capable of communicating via communication connections. Therefore, the electronic device 700 can operate in a networked environment using logical connections to one or more other servers, network personal computers (PCs), or another network node.

[0111] Input device 750 can be one or more input devices, such as a mouse, keyboard, trackball, etc. Output device 760 can be one or more output devices, such as a monitor, speaker, printer, etc. Electronic device 700 can also communicate with one or more external devices (not shown) via communication unit 740 as needed. These external devices include storage devices, display devices, etc., and can communicate with one or more devices that enable user interaction with electronic device 700, or with any device that enables electronic device 700 to communicate with one or more other electronic devices (e.g., network card, modem, etc.). Such communication can be performed via input / output (I / O) interface (not shown).

[0112] According to an exemplary implementation of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions thereon, wherein the computer-executable instructions are executed by a processor to implement the methods described above. According to an exemplary implementation of this disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, which are executed by a processor to implement the methods described above.

[0113] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, devices, and computer program products implemented according to this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0114] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0115] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0116] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0117] Various implementations of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed implementations. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to technology in the market, or to enable others skilled in the art to understand the various implementations disclosed herein.

Claims

1. An audio processing method, comprising: Construct structured data associated with the audio data stream; Based on the structured data, multiple voiceprint clustering results are generated, and the multiple voiceprint clustering results correspond to different numbers of voiceprint tags; as well as Based on the evaluation information corresponding to the various voiceprint clustering results, voiceprint information associated with the audio data stream is determined, wherein the voiceprint information indicates the voiceprint feature representation and voiceprint label of at least one speaker in the audio data stream.

2. The method according to claim 1, wherein, The construction of structured data associated with the audio data stream includes: Obtain the voiceprint matrix corresponding to the audio data stream, wherein the voiceprint matrix includes a first number of dimensions; and The voiceprint matrix is ​​mapped to the structured data, which includes a second number of dimensions, and the second number is less than the first number.

3. The method according to claim 2, wherein the voiceprint matrix includes a first matrix, and the step of obtaining the voiceprint matrix corresponding to the audio data stream includes: Determine at least one preset voiceprint feature representation; as well as First perturbation information is added to the at least one preset voiceprint feature representation to construct the first matrix.

4. The method according to claim 2, wherein the voiceprint matrix includes a second matrix, and the step of obtaining the voiceprint matrix corresponding to the audio data stream includes: Obtain a historical voiceprint set, which includes a mapping between the voiceprint tags and voiceprint feature representations of at least one historical speaker; as well as Based on the historical voiceprint set, the second matrix is ​​constructed.

5. The method according to claim 2, wherein the voiceprint matrix includes a third matrix, and the audio data stream includes multiple segments arranged in chronological order, wherein obtaining the voiceprint matrix corresponding to the audio data stream includes: Based on a preset window length, multiple consecutive voiceprint feature representations are determined from the audio data stream, including the voiceprint feature representation corresponding to the current moment. as well as The third matrix is ​​constructed based on the multiple continuous voiceprint feature representations.

6. The method according to claim 2, wherein the voiceprint matrix includes a fourth matrix, and the step of obtaining the voiceprint matrix corresponding to the audio data stream includes: A second perturbation information is added to the voiceprint feature representation corresponding to the current segment of the audio data stream to construct the fourth matrix.

7. The method according to claim 1, wherein, Based on the structured data, various voiceprint clustering results are generated, including: Determine multiple clustering parameters, which indicate the number of speakers in the audio data stream; The structured data is clustered based on the multiple clustering parameters to obtain multiple sets of voiceprint tags corresponding to the multiple cluster numbers; and Based on the multiple voiceprint tag sets, various voiceprint clustering results are constructed.

8. The method of claim 7, wherein the clustering parameters include at least one of the following: The first parameter indicates at least one preset number of speakers in the audio data stream; The second parameter indicates the number of speakers in the historical audio data stream.

9. The method according to claim 1, wherein, The step of determining the voiceprint information associated with the audio data stream based on the evaluation information corresponding to the various voiceprint clustering results includes: Determine multiple evaluation information corresponding to the various voiceprint clustering results; Based on the multiple evaluation information, the target number of speakers in the audio data stream is determined; and Based on the voiceprint clustering results corresponding to the target number, voiceprint information associated with the audio data stream is determined.

10. The method according to claim 9, wherein, Determining the target number of speakers in the audio data stream based on the multiple evaluation information includes: From the plurality of evaluation information, a first evaluation information and a second evaluation information obtained based on the number of speakers in the historical audio data stream are determined, wherein the first evaluation information is superior to the second evaluation information; The first evaluation information is updated based on a preset update coefficient; and Based on the first evaluation information and the second evaluation information, the target number of speakers in the audio data stream is determined.

11. The method according to claim 10, wherein, Determining the target number of speakers in the audio data stream based on the first evaluation information and the second evaluation information includes: In response to the first evaluation information being superior to the second evaluation information, the target number of speakers in the audio data stream is determined based on the clustering parameters corresponding to the first evaluation information; and In response to the first evaluation information being no better than the second evaluation information, the number of speakers in the historical audio data stream is determined as the target number of speakers in the audio data stream.

12. The method according to claim 1, further comprising: Obtain the historical voiceprint set, as well as the voiceprint feature representation and voiceprint label corresponding to at least one current speaker in the current segment of the audio data stream; Based on the voiceprint feature representation and voiceprint label corresponding to the current speaker, determine the target label matched by the current speaker; Based on the historical voiceprint set, as well as the voiceprint feature representation and voiceprint label corresponding to the current speaker, multiple correlations between the current speaker and historical speakers are determined. as well as Based on the aforementioned multiple correlations, the historical voiceprint set and the target label are updated.

13. The method according to claim 12, wherein, The step of updating the historical voiceprint set and the target tag based on the multiple correlations includes: In response to the fact that the matching relationship between the current speaker and the historical speaker corresponding to the target tag does not meet the first preset condition, the historical voiceprint set and the target tag are updated based on multiple correlations between the current speaker and the historical speaker; In response to the matching relationship between the current speaker and the historical speaker corresponding to the target tag satisfying a first preset condition, the historical voiceprint set is updated based on multiple correlations between the current speaker and the historical speaker.

14. The method according to claim 13, wherein, The step of responding to the situation where the matching relationship between the current speaker and the historical speakers corresponding to the target tag does not meet the first preset condition, and updating the historical voiceprint set and the target tag based on the correlation between the current speaker and the historical speakers, includes: In response to the fact that the matching relationship between the current speaker and the historical speaker corresponding to the target tag does not meet the first preset condition, and the correlation between the current speaker and the historical speaker corresponding to the target tag does not exceed the first threshold, the target tag is updated; The historical voiceprint set is updated based on the target label and the voiceprint feature representation of the current speaker.

15. The method according to claim 13, wherein, The step of responding to the situation where the matching relationship between the current speaker and the historical speakers corresponding to the target tag does not meet the first preset condition, and updating the historical voiceprint set and the target tag based on the correlation between the current speaker and the historical speakers, includes: In response to the fact that the matching relationship between the current speaker and the historical speaker corresponding to the target tag does not meet the first preset condition, and the correlation between the current speaker and the historical speaker corresponding to the target tag exceeds the second threshold, the target tag is updated based on the voiceprint tags in the historical voiceprint set; The historical voiceprint set is updated based on the target label and the voiceprint feature representation of the current speaker.

16. The method according to claim 15, wherein updating the historical voiceprint set based on the target label and the voiceprint feature representation of the current speaker comprises: In response to the fact that the number of voiceprint feature representations corresponding to the target tag in the historical voiceprint set exceeds the preset storage number, the voiceprint density corresponding to the target tag is determined. Based on the voiceprint density, edge feature representations are determined from the voiceprint feature representations corresponding to the target label; The historical voiceprint set is updated based on the edge feature representation and the voiceprint feature representation of the current speaker.

17. An apparatus for information processing, comprising: The building module is configured to construct structured data associated with the audio data stream; The generation module is configured to generate multiple voiceprint clustering results based on the structured data, wherein the multiple voiceprint clustering results correspond to different numbers of voiceprint tags; as well as The determination module is configured to determine voiceprint information associated with the audio data stream based on the evaluation information corresponding to the multiple voiceprint clustering results, wherein the voiceprint information indicates the voiceprint feature representation and voiceprint label of at least one speaker in the audio data stream.

18. An electronic device comprising: At least one processor; as well as At least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor, the instructions causing the electronic device to perform the method according to any one of claims 1 to 16 when executed by the at least one processor.

19. A computer-readable storage medium having stored thereon computer-executable instructions that can be executed by a processor to implement the method according to any one of claims 1 to 16.

20. A computer program product tangibly stored in a computer storage medium and comprising computer-executable instructions that, when executed by a device, cause the device to perform the method according to any one of claims 1 to 16.

Images