Call audio playing method and device and computing equipment

Abstract

The embodiment of the invention discloses a call audio playing method and device and computing equipment, and belongs to the technical field of Internet. The method comprises the steps that audio data of a plurality of second accounts in a conference call are acquired, and the audio data of at least one second account comprises a plurality of personal audio. And performing noise reduction processing on the audio data of each second account to obtain noise-reduced audio data corresponding to each second account. And carrying out human voice separation on the audio data after noise reduction processing corresponding to each second account to obtain at least one human voice audio corresponding to each second account. And determining a target human voice audio needing to be played in the at least one human voice audio corresponding to each second account, and playing the target human voice audio. By adopting the method and the device, noise reduction processing can be performed on the call audio of the conference call, and the voice audio in the conference call can be selectively played, so that the quality of the call audio can be improved.

Description

[0001] This application claims priority to Chinese patent application No. 202311047303.4, filed on August 18, 2023, and entitled “A method and device for speech separation”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of Internet technology, and in particular to a method, device and computing equipment for playing call audio. Background Art

[0003] With the development of Internet technology, conference calls have become a common form of communication.

[0004] Since each participant is in a different environment when participating in a conference call, various environmental sounds may appear in the call audio of the conference call, thereby reducing the quality of the call audio. Summary of the invention

[0005] The embodiments of the present application provide a method, apparatus and computing device for playing call audio, which can improve the quality of call audio. The corresponding technical solutions are as follows:

[0006] In a first aspect, a method for playing call audio is provided, the method comprising: in response to a conference call joining operation of a first account, joining the first account to a conference call. Obtaining audio data of each second account in the conference call except the first account, wherein the audio data of at least one second account includes multiple human voice audios. Performing noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account. Performing human voice separation on the noise-reduced audio data corresponding to each second account to obtain at least one human voice audio corresponding to each second account. Determining the target human voice audio to be played in the at least one human voice audio corresponding to each second account. Playing the target human voice audio corresponding to each second account.

[0007] Among them, the method for playing the call audio can be executed by the terminal logged in by the first account, or by the server that forwards the call audio of the conference call. In the scheme shown in the present application, for the audio data of each second account, the audio data can be first subjected to noise reduction processing, and then the audio data after noise reduction processing can be separated from the human voice. Then, the human voice audio can be selectively played in at least one human voice audio corresponding to each second account. In this way, the call audio of the conference call only includes human voice audio, the ambient sound of the environment in which the participants are located can be filtered out, and only the human voice audio is retained, thereby improving the quality of the call audio.

[0008] In one achievable manner, the audio data of each second account is subjected to noise reduction processing to obtain the noise-reduced audio data corresponding to each second account, including: for the audio data of each second account, extracting the audio features corresponding to the audio data. Inputting the audio features into the first speech detection algorithm to obtain the first probability that each audio frame in the audio data belongs to human voice audio. Inputting the audio features into the second speech detection algorithm to obtain the second probability that each audio frame in the audio data belongs to human voice audio. The first probability and the second probability that each audio frame belongs to human voice audio are merged to obtain the third probability that each audio frame belongs to human voice audio. Based on the third probability that each audio frame belongs to human voice audio, the audio data is subjected to noise reduction processing to obtain the noise-reduced audio data.

[0009] In the solution provided in the present application, when performing noise reduction processing on audio data, the noise reduction processing on the audio data can be performed by fusing two different speech detection algorithms, which can improve the ability to perform noise reduction processing on the audio data.

[0010] In an achievable manner, determining the target human voice audio to be played from the at least one human voice audio corresponding to each second account includes: for the at least one human voice audio corresponding to each second account, determining the human voice audio of the participant from the at least one human voice audio, and determining the human voice audio of the participant as the target human voice audio to be played. In this way, the call audio of the conference call played can only include the human voice audio of the participant, which can improve the quality of the call audio.

[0011] In an achievable manner, determining the human voice audio of the participant in at least one human voice audio includes: when there are multiple human voice audios, determining the average loudness corresponding to the multiple human voice audios, and determining the human voice audio of the participant whose average loudness is greater than that of other human voice audios in the multiple human voice audios as the human voice audio of the participant. In this way, the human voice audio of the participant can be identified in the multiple human voice audios only by loudness detection, which can improve the efficiency of identifying the human voice audio of the participant.

[0012] In one achievable manner, determining the human voice audio of the participant in at least one human voice audio includes: randomly playing a human voice audio in the at least one human voice audio. If the human voice audio being played is the human voice audio of a non-participant, then in response to an audio replacement instruction triggered by the first account to the second account, replacing the human voice audio being played in the at least one human voice audio. In this way, the user can customize the human voice audio to be played, which can improve the flexibility of the call audio playback.

[0013] In one achievable method, determining the human voice audio of a participant in at least one human voice audio includes: determining whether there is human voice audio in at least one human voice audio whose voiceprint similarity with the participant audio in the participant audio library is greater than a similarity threshold, the participant audio corresponding to each participant who joins the conference call is stored in the participant audio, and if there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

[0014] In the solution provided in the present application, the human voice audio separated from the call audio can be identified through a pre-established participant audio library, and then the identified human voice audio can be selectively played, thereby improving the flexibility of call audio playback.

[0015] In an achievable manner, the method further includes: determining and displaying a participant ID corresponding to the participant's voice audio, and pausing the playback of the voice audio corresponding to the specified participant in response to a participant blocking instruction triggered by the first account and carrying the participant ID corresponding to the specified participant. In this way, the user can selectively play the voice audio in the call audio, which can improve the flexibility of the call audio playback.

[0016] In one achievable manner, a conference call includes multiple conference topics, and a target voice audio to be played is determined from at least one voice audio corresponding to each second account, including: determining the conference topic currently being played in the conference call, and determining the target voice audio to be played from at least one voice audio corresponding to each second account participating in the currently being played conference topic. In this way, the voice audio in the call audio can be selectively played according to the conference topic being played in the conference call, which can improve the flexibility of the call audio playback.

[0017] In a second aspect, a device for playing call audio is provided, the device comprising:

[0018] A joining module, configured to join the first account to the conference call in response to the conference call joining operation of the first account;

[0019] An acquisition module, configured to acquire audio data of a plurality of second accounts, wherein the audio data of at least one second account includes a plurality of human voice audios;

[0020] A noise reduction module, used to perform noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account;

[0021] A separation module, used to perform human voice separation on the noise reduction processed audio data corresponding to each second account, to obtain at least one human voice audio corresponding to each second account;

[0022] A determination module, configured to determine a target human voice audio to be played from at least one human voice audio corresponding to each second account;

[0023] The playing module is used to play the target human voice audio corresponding to each second account.

[0024] In one achievable manner, the noise reduction module is used to: for the audio data of each second account, extract the audio features corresponding to the audio data. Input the audio features into the first voice detection algorithm to obtain a first probability that each audio frame in the audio data belongs to human voice audio, input the audio features into the second voice detection algorithm to obtain a second probability that each audio frame in the audio data belongs to human voice audio. The first probability and the second probability that each audio frame belongs to human voice audio are merged to obtain a third probability that each audio frame belongs to human voice audio, and based on the third probability that each audio frame belongs to human voice audio, the audio data is subjected to noise reduction processing to obtain audio data after noise reduction processing.

[0025] In one achievable manner, the determination module is used to: for at least one human voice audio corresponding to each second account, determine the human voice audio of the participant in the at least one human voice audio, and determine the human voice audio of the participant as the target human voice audio to be played.

[0026] In one achievable manner, the determination module is used to: when there are multiple human voice audios, determine the average loudness corresponding to the multiple human voice audios respectively; and determine the human voice audio among the multiple human voice audios whose average loudness is greater than that of other human voice audios among the multiple human voice audios as the human voice audio of the participant.

[0027] In one achievable manner, a determination module is used to: randomly play a vocal audio in at least one vocal audio; if the played vocal audio is the vocal audio of a non-participant, then in response to an audio change instruction triggered by the first account to the second account, replace the played vocal audio in at least one vocal audio.

[0028] In one achievable manner, a determination module is used to: determine, in at least one human voice audio, whether there is human voice audio whose voiceprint similarity with the participant audio corresponding to the participant audio in the participant audio library is greater than a similarity threshold, the participant audio corresponding to each participant who joins the conference call is stored in the participant audio, and if there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

[0029] In one achievable method, the device also includes a shielding module, which is used to: determine and display the participant identification corresponding to the participant's voice audio, and in response to a participant shielding instruction triggered by the first account and carrying the participant identification corresponding to the specified participant, pause the playback of the voice audio corresponding to the specified participant.

[0030] In one achievable manner, a conference call includes multiple conference topics, and a determination module is used to: determine the conference topic currently being conducted in the conference call, and determine the target voice audio that needs to be played from at least one voice audio corresponding to each second account participating in the currently ongoing conference topic.

[0031] In a third aspect, a computing device cluster is provided, including at least one computing device, each computing device including a processor and a memory. The processor of at least one computing device is used to execute instructions stored in the memory of at least one computing device, so that the computing device cluster executes the method described in the first aspect and / or any achievable manner of the first aspect.

[0032] According to a fourth aspect, a computer program product comprising instructions is provided. When the instructions are executed by a computing device, the computing device executes the method as described in the first aspect and / or any implementable manner of the first aspect.

[0033] In a fifth aspect, a computer-readable storage medium is provided, comprising computer program instructions. When the computer program instructions are executed by a computing device, the computing device executes the method described in the above-mentioned first aspect and / or any achievable method in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of a method for playing call audio provided in an embodiment of the present application;

[0035] Figure 2 It is a schematic diagram of the structure of a deep neural network provided in an embodiment of the present application;

[0036] Figure 3 This is a schematic diagram of a conference call provided by an embodiment of the present application;

[0037] Figure 4 This is a schematic diagram of a conference call provided by an embodiment of the present application;

[0038] Figure 5 It is a schematic diagram of an interface of a conference application provided in an embodiment of the present application;

[0039] Figure 6 This is a schematic diagram of a conference call provided by an embodiment of the present application;

[0040] Figure 7 It is a schematic diagram of an interface of a conference application provided in an embodiment of the present application;

[0041] Figure 8 This is a schematic diagram of a conference call provided by an embodiment of the present application;

[0042] Fig. 9 It is a structural schematic diagram of a call audio playback device provided in an embodiment of the present application;

[0043] Fig.10 is a schematic diagram of the structure of a computing device provided in an embodiment of the present application;

[0044] Fig.11 is a schematic diagram of the structure of a computing device cluster provided in an embodiment of the present application;

[0045] Fig.12 It is a structural diagram of a computing device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0047] Conference calling is a way of communicating with multiple people online through a conference calling program. Since conference calling is not restricted by time, location and other factors, it is increasingly used in life and work.

[0048] In one example, a user can log in to his or her account in a conference call program running on a terminal, and then can operate the conference call program to create or join a conference call in the conference call program. When a user creates a meeting in the conference call program, the conference call program can send a conference creation request to the server through the terminal. After receiving the conference creation request sent by the terminal, the server can create a meeting in the server and generate a conference identifier (Identity document, ID) for the meeting, and then send the conference ID to the terminal. The user can forward the conference ID to other users who need to join the conference call. Other users can operate the conference call program, enter the received conference ID in the conference call program, and then join the corresponding conference call. Among them, each user who joins the conference call can be called a participant.

[0049] After a participant joins a conference call through the conference call program running on the terminal, if the participant operates the conference call program to enable the microphone permission of the terminal, the conference call program can collect the participant's audio through the microphone of the terminal and then send the collected audio to the server. During the conference call, the participant who joins the conference call can receive and play the audio of other participants sent by the server through the terminal, so that multiple participants can conduct an online conference call.

[0050] Since each participant may be in a different environment when participating in a conference call, such as at home, in a conference room, or at a station, the audio of the participant collected by the terminal may include ambient sound and non-participant voice audio, which results in a complex ambient sound and non-participant voice audio received by each participant through the terminal, which in turn reduces the quality of the call audio and affects the participant's experience of the conference call. In addition, it is possible that multiple participants join the conference call through the same terminal, so other participants participating in the conference call may not be able to distinguish between the multiple participants.

[0051] The call audio playback method provided in the embodiment of the present application can filter out noise in the call audio and distinguish the human voice audio of each participant, thereby improving the quality of the call audio and improving the experience of the participants in the conference call.

[0052] Figure 1 This is a flow chart of a method for playing call audio provided by an embodiment of the present application. The method can be executed by a terminal that joins a conference call through a first account, or by a server that forwards the call audio. Alternatively, the method is partially executed by the terminal and partially executed by the server, for example, the server executes the following steps 101 to 104, and the terminal executes the following steps 105-106. Figure 1 , the method for playing the call audio includes:

[0053] Step 101: In response to a conference call joining operation of a first account, the first account is added to the conference call.

[0054] In the case where step 101 is performed by the terminal, the terminal may run a conference call program, and the account logged in to the conference call program is the first account. The participant may operate the conference call program to join the corresponding conference call, for example, the participant may operate the conference call program to create and join the corresponding conference call, or join the corresponding conference call by entering a conference ID. In response to the conference call joining operation triggered by the participant in the conference call program, the terminal may send a conference joining request carrying the account identifier corresponding to the first account to the server, so that the server joins the first account to the conference call after receiving the conference joining request.

[0055] When step 101 is performed by the server, the conference call joining operation corresponding to the first account may be a conference joining request sent by the receiving terminal and carrying the account identifier corresponding to the first account. After receiving the conference joining request carrying the account identifier corresponding to the first account, the server may join the first account to the conference call.

[0056] Among them, in the embodiments of the present application, any application with a multi-person voice call function can be called a conference call program, for example, it can be enterprise office software, video chat software, or video live broadcast software, etc.

[0057] The following steps 102 to 106 are described in detail by taking the execution of the terminal logging into the first account as an example:

[0058] Step 102: Obtain audio data of multiple second accounts, wherein the audio data of at least one second account includes multiple human voice audios.

[0059] After the first account joins the conference call, the server can send the audio data of each second account in the conference call except the first account to the terminal logged in by the first account. The audio data corresponding to each of the first account and the second account is collected by the microphone of the terminal logged in by each account, and the number of the terminal microphones can be one or more. The audio data of each second account sent by the server is a different audio frame sequence.

[0060] The audio data of at least one second account may include multiple human voice audios, which may include human voice audios of one or more participants, and may also include human voice audios of one or more non-participants. For example, multiple participants participate in a conference call in a conference room by logging into a terminal of a second account. For another example, a participant participates in a company conference call at home by logging into a terminal of the second account. The terminal may collect human voice audios of the participant's family members, and the participant's family members are non-participants.

[0061] Step 103: Perform noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account.

[0062] After receiving the audio data corresponding to each second account, the terminal logged in to the first account can first filter out the noise and the like included in the audio data corresponding to each second account through a noise reduction algorithm to obtain the noise-reduced audio data corresponding to each second account.

[0063] In one example, performing noise reduction processing on the audio data of each second account may include: for the audio data of each second account, extracting audio features corresponding to the audio data. Inputting the audio features into the first speech detection algorithm to obtain a first probability that each audio frame in the audio data belongs to human voice audio. Inputting the audio features into the second speech detection algorithm to obtain a second probability that each audio frame in the audio data belongs to human voice audio. Fusion of the first probability and the second probability that each audio frame belongs to human voice audio to obtain a third probability that each audio frame belongs to human voice audio. Based on the third probability that each audio frame belongs to human voice audio, performing noise reduction processing on the audio data to obtain audio data after noise reduction processing.

[0064] Among them, the first voice detection algorithm and the second voice detection algorithm belong to two different voice detection algorithms. In implementation, the two voice detection algorithms can be used to respectively determine the first probability sequence composed of the first probability that each audio frame in the audio data corresponding to the second account belongs to human voice audio, and the second probability sequence composed of the second probability that each audio frame belongs to human voice audio. The value range of the probability value in the first probability sequence and the second probability sequence is [0, 1]. After obtaining the first probability sequence and the second probability sequence, the first probability sequence and the second probability sequence can be fused to obtain a third probability sequence composed of the third probability that each audio frame belongs to human voice audio. Then, the audio data of the second account is subjected to noise reduction processing according to the obtained third probability sequence. The fusion of the first probability sequence and the second probability sequence can be multiplying the first probability and the second probability corresponding to the same audio frame, or binarizing one of the first probability or the second probability, and then multiplying the probability that is not binarized in the first probability or the second probability, and then obtaining the third probability. In this way, in the embodiment of the present application, the noise reduction processing of the audio data by two different voice detection algorithms can improve the effect of noise reduction on the audio data.

[0065] Step 104: perform human voice separation on the noise reduction processed audio data corresponding to each second account to obtain at least one human voice audio corresponding to each second account.

[0066] For each second account, after obtaining the audio data of the second account after performing noise reduction processing on the audio data, the terminal can separate at least one human voice audio included in the second account through a human voice separation algorithm.

[0067] Step 105: Determine the target human voice audio to be played from the at least one human voice audio corresponding to each second account.

[0068] For each second account, after obtaining at least one human voice audio corresponding to each second account, the terminal can determine the target human voice audio that needs to be played in the at least one human voice audio. In one example, each human voice audio of each separated second account can be identified to obtain the human voice audio belonging to the participant, and then the human voice audio of the participant is determined as the human voice audio that needs to be played. In another example, the user who operates the terminal corresponding to the first account can listen to the human voice audio of the specified participant. Therefore, the identity of the participant corresponding to each human voice audio can also be identified, and then the human voice audio belonging to the specified participant is determined as the human voice audio that needs to be played.

[0069] Step 106: Play the target human voice audio corresponding to each second account.

[0070] After determining the target human voice audio to be played in each human voice audio corresponding to each second account, the determined target human voice audio can be played. Since the played target human voice audio only eliminates the noise and noise, and only includes the human voice audio of the participant, the quality of the played call audio can be improved, providing the participant with a better experience of participating in the conference call.

[0071] When the terminal performs the processing of steps 102 to 106, it may receive a segment of audio data from each second account, and the duration of the segment of audio data may be several milliseconds, for example, 10 milliseconds. This can reduce the duration of the terminal processing the audio data of the second account, thereby not affecting the normal communication of multiple participants in the conference call.

[0072] In addition, the server performs the above steps 102 to 106 in a process that is basically the same as the terminal's execution process, and will not be repeated in the embodiment of the present application. In addition, when the server performs step 106, the target voice audio may be played by sending the determined target voice audio to the terminal logged in by the first account, so that the terminal logged in by the first account plays the target voice audio.

[0073] The embodiment of the present application also provides a deep neural network for performing the processing of the above steps 103 and 104. Figure 2 Schematic diagram of the structure of the deep neural network provided in the embodiment of the present application. Figure 2 As shown, the deep neural network can be composed of a feature extraction unit, an audio noise reduction unit and an audio separation unit. The feature extraction unit can be used to extract audio features corresponding to the audio data of the second account, the audio noise reduction unit can be used to perform noise reduction processing on the audio data of each second account, and the audio separation unit can be used to perform human voice separation on the noise-reduced audio data corresponding to each second account.

[0074] The feature extraction unit may include a conv layer and a Gru layer, wherein the conv layer is a convolution layer, the Gru layer is a gated recurrent unit (GRU), and the output of the conv layer is the input of the Gru layer. Before the audio data of the second account is input into the feature extraction unit, the audio data may be first converted into the frequency domain using a short-time Fourier transform (STFT) to obtain the frequency domain data corresponding to the audio data. After obtaining the frequency domain data corresponding to the audio data, the frequency domain data may be input into the conv layer for convolution processing, and then the output of the conv layer may be input into the Gru layer for further feature extraction, thereby obtaining the audio features corresponding to the audio data of the second account.

[0075] The audio noise reduction unit may include two detection subunits connected in parallel and a detection subunit connected in series with the two detection subunits. One of the detection subunits is composed of a linear layer and a mask layer connected in series, and the other detection subunit is composed of a linear layer and a vad layer connected in series. The linear layer included in the two detection subunits is a fully connected layer, which can further extract features of the input audio features. Different speech detection algorithms can be set in the mask layer and the vad layer. For example, the vad layer is provided with a voice activity detection algorithm (Voice Activity Detection, VAD).

[0076] In one example, the vad layer can output a first probability sequence that each audio frame of the audio data corresponding to the second account belongs to human voice audio, and the mask layer can output a second probability sequence that each audio frame of the audio data corresponding to the second account belongs to human voice audio. The first probability sequence and the second probability sequence output by the vad layer and the mask layer can be input to the denoising subunit. The denoising subunit can fuse the first probability sequence and the second probability sequence to obtain a third probability sequence. In addition, the input of the denoising subunit can also include frequency domain data corresponding to the audio data processed by the short-time Fourier transform (STFT) in the feature extraction unit. The denoising subunit can perform denoising processing on the frequency domain data according to the third probability sequence to obtain the frequency domain data after denoising processing.

[0077] The audio separation unit may include a conv layer and a decoder layer, wherein the conv layer is a convolutional layer, and the decoder layer includes a decoder. The input of the conv layer is the frequency domain data after noise reduction processing, and the output of the conv layer is the input of the decoder. The conv layer is used to extract features from the frequency domain data to obtain frequency domain feature data. The decoder layer is used to decode the frequency domain feature data, and then decode at least one human voice audio sequence. Among them, the human voice audio sequence decoded by the decoder layer is frequency domain data, so the human voice audio sequence decoded by the decoder layer can be converted into time domain data, and then a human voice audio sequence in the time domain can be obtained.

[0078] In one example, the decoder layer can decode a specified number of human voice audio sequences, and each human voice audio sequence can output the corresponding confidence level of the human voice audio sequence. In implementation, the human voice audio sequence whose corresponding confidence level is greater than the confidence level threshold can be determined as the human voice audio sequence separated from the audio data corresponding to the second account. In addition, the above-mentioned deep neural network can be pre-trained with sample data before application, such as supervised training, and its training process can be implemented using relevant training methods, which will not be repeated in the embodiments of the present application.

[0079] In the call audio playback method provided in the embodiment of the present application, the processing of step 105 may include: for at least one human voice audio corresponding to each second account, determining the human voice audio of the participant in the at least one human voice audio, and determining the human voice audio of the participant as the target human voice audio to be played.

[0080] In an embodiment of the present application, at least one human voice audio corresponding to the obtained second account is separated after noise reduction processing, so each human voice audio only includes the speaking voice of one person and does not include environmental sound. On this basis, by identifying whether each separated human voice audio belongs to the voice of a participant participating in the conference call, the human voice audio belonging to the participant is determined as the target human voice audio for playback. In this way, the call audio corresponding to the conference call played by the first account terminal only includes the human voice audio of the participants participating in the conference call, so the first account terminal can avoid playing the noise in the environment of other participants, the voices of non-participants, etc., thereby improving the quality of the call audio and improving the experience of the participants in the conference call.

[0081] The following describes several methods for determining the human voice audio of a participant in at least one human voice audio provided in the embodiments of the present application:

[0082] Processing 1: When there are multiple human voice audios, determine the average loudness corresponding to the multiple human voice audios, and determine the human voice audio among the multiple human voice audios whose average loudness is greater than that of other human voice audios among the multiple human voice audios as the human voice audio of the participant.

[0083] During the conference call, the participants are generally close to the microphone of the terminal and speak at a high frequency. Therefore, in the audio collected by the terminal, the human voice audio of the participant should be the loudest and the longest duration. Therefore, for multiple human voice audio separated from the audio data corresponding to a second account, the human voice audio with the loudest continuous loudness in the multiple human voice audios can be detected, and then the human voice audio with the loudest continuous loudness can be determined as the human voice audio of the participant. For example, a detection time window can be set, and then the average loudness of the audio frame of each human voice audio in the detection time window can be detected, and the human voice audio with the loudest continuous loudness is the human voice audio with the loudest continuous loudness in the multiple human voice audios. In an example, the human voice audio can be Fourier transformed to obtain the frequency domain data corresponding to the human voice audio. For each audio frame, the amplitude of the audio frame in the frequency domain data can be determined as the loudness of the audio frame, or the amplitude of the audio frame in the frequency domain data can be determined as the loudness of the audio frame. Among them, the specified frequency segment can be a human voice frequency segment, which can further eliminate the influence of other noises in the human voice audio on the loudness calculation, and can improve the accuracy of the human voice loudness calculation.

[0084] Figure 3 is a schematic diagram of a conference call provided in an embodiment of the present application, such as Figure 3 As shown, the participants corresponding to the second account in the conference call may include participant A and participant B. Among them, the second account logged in by participant A is the second account a, and the second account logged in by participant B is the second account b. Microphone A is the microphone corresponding to the terminal used by participant A to participate in the conference call, and microphone B is the microphone corresponding to the terminal used by participant B to participate in the conference call. The audio collected by microphone A includes the human voice audio of participant A and the human voice audio of non-participants. For example, if participant A participates in the conference call at home, the human voice audio of non-participants can be the human voice audio of family members of participant A. The audio collected by microphone B includes the human voice audio of participant B.

[0085] See also Figure 3After receiving the call audio, the terminal logged in to the first account can perform noise reduction processing on the audio data corresponding to the second account a and the second account b respectively, and then perform voice separation processing on the noise-reduced audio data corresponding to the second account a and the second account b respectively. For the second account a, the audio sequence of participant A and the audio sequence of non-participants can be separated, and for the second account b, the audio sequence of participant B can be separated. Since the second account a includes two human voice audio sequences, the human voice audio sequence with the largest continuous loudness can be determined in these two human voice audio sequences as the human voice audio sequence of the participant. In this way, the terminal logged in to the first account can eliminate the human voice audio sequence of non-participants and only include the audio sequences of participants A and B.

[0086] Figure 4 is a schematic diagram of a conference call provided by an embodiment of the present application, and Figure 3 same, Figure 4 The participants corresponding to the second account in the conference call may include participant A and participant B. The second account logged in by participant A is second account a, and the second account logged in by participant B is second account b. Microphone A is the microphone corresponding to the terminal used by participant A to participate in the conference call, and microphone B is the microphone corresponding to the terminal used by participant B to participate in the conference call. Since participant A and participant B are physically close, such as in the same conference room, microphone A will also collect the human voice audio of participant B, and microphone B will also collect the human voice audio of participant A, which will cause the audio to overlap with each other, reducing the quality of the call audio.

[0087] See also Figure 4 , after receiving the call audio, the terminal logged in to the first account can perform noise reduction processing on the audio data corresponding to the second account a and the second account b respectively, and then perform human voice separation processing on the noise-reduced audio data corresponding to the second account a and the second account b respectively. For the second account a, the audio sequences of participants A and B can be separated, and for the second account b, the audio sequences of participants A and B can also be separated. Participant A is closer to microphone A than participant B. Therefore, in the audio sequences of participants A and B separated from the second account a, the continuous loudness of the audio sequence of participant A is greater than that of participant B. Therefore, the audio sequence of participant A separated from the second account a can be determined as the audio sequence of the participant. Similarly, the audio sequence of participant B separated from the second account a can be determined as the audio sequence of the participant. In this way, the terminal logged in to the first account can eliminate the superimposed human voice audio, thereby improving the quality of the call audio.

[0088] Processing 2: Randomly play a vocal audio in at least one vocal audio. If the played vocal audio is the vocal audio of a non-participant, then in response to the audio change instruction triggered by the first account to the second account, replace the played vocal audio in at least one vocal audio.

[0089] If the noise reduction processed audio data corresponding to the second account separates multiple human voice audios, the terminal can randomly play the human voice audios among the multiple human voice audios. If the user determines that the randomly played human voice audios among the multiple human voice audios do not belong to the human voice audios of the participant, the conference call program can be operated to replace the human voice audios played among the multiple human voice audios until the user determines that the human voice audios played among the multiple human voice audios belong to the human voice audios of the participant.

[0090] like Figure 3 As shown, after separating the audio sequence of participant A and the audio sequence of non-participants from the audio data of the second account a, the terminal logged in to the first account can randomly select an audio sequence from the audio sequence of participant A and the audio sequence of non-participants to play. If the audio sequence played is the audio sequence of non-participants, the user can operate the conference call program to replace the audio of the non-participants being played with the audio sequence of participant A.

[0091] Figure 5 This is a schematic diagram of an interface of a conference application provided by an embodiment of the present application. The account identifier corresponding to the second account of the participating conference call can be displayed in the interface of the conference application. For example, the account identifier is the employee number of the participant in the enterprise. If multiple human voice audios are separated from the audio data corresponding to the second account, an audio playback switching control is displayed at a position corresponding to the account identifier of the second account in the interface of the conference application. The user can click on the audio playback switching control to change the human voice audio played among the multiple human voice audios corresponding to the second account.

[0092] Processing three: in at least one human voice audio, determine whether there is human voice audio whose voiceprint similarity with the participant audio in the participant audio library is greater than the similarity threshold, and the participant audio library stores the participant audio corresponding to each participant who joins the conference call; if there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

[0093] For participants in a conference call, a participant audio library corresponding to the participant can be established in advance. For example, the conference application belongs to the office application of the enterprise, and the participant audio library can store audio data corresponding to each employee in the enterprise, such as voiceprint data. When multiple employees join the conference call, the server can obtain the voiceprint data of the multiple employees from the participant audio library and send it to each terminal that joins the conference call. After separating the at least one human voice audio corresponding to the second account, the terminal can respectively match the at least one human voice audio with the voiceprint similarity corresponding to the voiceprint data of the multiple employees. If it is determined that there is a human voice audio in at least one human voice audio whose voiceprint similarity with the voiceprint data of any employee is greater than the set similarity threshold, it can be determined that the human voice audio is the human voice audio of the participant, and the participant information of the participant can be determined, and then it can be displayed in the interface of the conference application.

[0094] Figure 6 is a schematic diagram of a conference call provided in an embodiment of the present application, such as Figure 6 As shown, the participants corresponding to the second account in the conference call may include participant A and participant B. The second account logged in by participant A is second account a, and the second account logged in by participant B is second account b. Microphone A is the microphone corresponding to the terminal used by participant A to participate in the conference call, and microphone B is the microphone corresponding to the terminal used by participant B to participate in the conference call. The audio collected by microphone A includes the human voice audio of participant A and the human voice audio of non-participants.

[0095] See also Figure 6 After receiving the call audio, the terminal logged in to the first account can perform noise reduction processing on the audio data corresponding to the second account a and the second account b respectively, and then perform voice separation processing on the noise-reduced audio data corresponding to the second account a and the second account b respectively. For the two audio sequences separated from the second account a, after comparing them with the middle voiceprint data in the participant audio library, it can be determined that one audio sequence is the audio sequence of participant A and the other audio sequence is the audio sequence of a non-participant. For the audio sequence separated from the second account b, after comparing it with the middle voiceprint data in the participant audio library, the separated audio sequence can be the audio sequence of participant B.

[0096] In one example, after determining the human voice audio of the participant according to the participant audio library, the participant ID corresponding to the human voice audio of the participant can be displayed on the terminal. If the terminal receives a participant blocking instruction triggered by the first account and carrying the participant ID corresponding to the specified participant, then in response to the participant blocking instruction, the human voice audio corresponding to the specified participant is paused.

[0097] Figure 7This is a schematic diagram of the interface of a conference application provided in an embodiment of the present application. In the interface of the conference application, the participant information of each participant to whom the voice audio corresponding to the second account of the participating conference call belongs, such as the employee number, name, etc. of the participant, can be displayed. Corresponding to the participant information display position, an option to turn on the participant audio can be displayed. For example, by default, this option can be turned on, that is, the terminal plays the voice audio of each participant by default. The user can choose to adjust the option to the off state, thereby triggering the participant blocking instruction corresponding to the option. After receiving the participant blocking instruction, the terminal can pause playing the corresponding participant's voice audio.

[0098] In one achievable manner, the conference topic currently being conducted in the conference call is determined, and a target voice audio to be played is determined from at least one voice audio corresponding to each second account participating in the currently being conducted conference topic.

[0099] For some large online conferences, the entire conference process may be divided into multiple conference topics. Multiple participants in a conference call may participate in different conference topics. Or the participating identities of different participants in each conference topic may be different, for example, participant A's participating identity in the first conference topic is a speaker, and the participating identity in the second conference topic is a listener. In one example, the creator of the conference call can pre-set the conference topics included in the conference call, as well as the second account, participants, etc. participating in each conference call. The creator of the conference call can also pre-set the duration of each conference topic in the conference call, or control the switching of conference topics during the conference call.

[0100] In the above step 105, in the process of determining the target human voice audio to be played in at least one human voice audio corresponding to the second account, the conference theme of the current conference call can be determined first, and then in each second account participating in the current conference theme (or in the second account of the speaker identity in the current conference theme), the target human voice audio to be played can be determined. In this way, the processing amount of audio data can be reduced and the processing efficiency can be improved.

[0101] Figure 8 is a schematic diagram of a conference call provided in an embodiment of the present application, such as Figure 8 As shown, the participants corresponding to the second account in the conference call may include participants A and B corresponding to conference topic 1, and participant C corresponding to conference topic 2. Participants A and B can participate in the conference call by logging into the terminal of the second account a, and participant C can participate in the conference call by logging into the terminal of the second account c. Microphone A is the microphone corresponding to the terminal used by participants A and B to participate in the conference call, and microphone B is the microphone corresponding to the terminal used by participant C to participate in the conference call.

[0102] See also Figure 8 , if the current conference call is in progress with conference topic 1, the terminal logged into the first account can receive the audio data corresponding to the second account a, and then the audio data corresponding to the second account a can be subjected to noise reduction and voice separation processing. Then, the two audio sequences separated from the second account a can be compared with the middle voiceprint data in the participant audio library, and it can be determined that one audio sequence is the audio sequence of participant A and the other audio sequence is the audio sequence of participant B. In this way, filtering at least one voice audio corresponding to the second account that needs to be matched by the conference topic can reduce the amount of recognition processing of the voice audio and improve processing efficiency.

[0103] Fig. 9 is a schematic diagram of a structure of a call audio playback device provided in an embodiment of the present application, such as Fig. 9 As shown, the device comprises:

[0104] The joining module 910 is configured to add the first account to the conference call in response to a conference call joining operation of the first account.

[0105] The acquisition module 920 is used to acquire audio data of multiple second accounts, wherein the audio data of at least one second account includes multiple human voice audios.

[0106] The noise reduction module 930 is used to perform noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account.

[0107] The separation module 940 is used to separate the human voice from the noise reduction processed audio data corresponding to each second account to obtain at least one human voice audio corresponding to each second account.

[0108] A determination module 950 is used to determine a target human voice audio to be played from the at least one human voice audio corresponding to each second account;

[0109] The playing module 960 is used to play the target human voice audio corresponding to each second account.

[0110] In one achievable manner, the noise reduction module 930 is used to:

[0111] For each audio data of the second account, extracting audio features corresponding to the audio data;

[0112] Inputting the audio feature into a first speech detection algorithm to obtain a first probability that each audio frame in the audio data belongs to human voice audio;

[0113] Inputting the audio feature into a second speech detection algorithm to obtain a second probability that each audio frame in the audio data belongs to human voice audio;

[0114] Merging the first probability and the second probability that each audio frame belongs to human voice audio to obtain a third probability that each audio frame belongs to human voice audio;

[0115] Based on the third probability that each audio frame belongs to human voice audio, the audio data is subjected to noise reduction processing to obtain audio data after noise reduction processing.

[0116] In one achievable manner, the determining module 950 is used to:

[0117] For at least one human voice audio corresponding to each second account, determine the human voice audio of the participant in the at least one human voice audio;

[0118] The human voice audio of the participant is determined as the target human voice audio to be played.

[0119] In one achievable manner, the determining module 950 is used to:

[0120] In a case where the number of the at least one human voice audio is plural, determining average loudnesses respectively corresponding to the plurality of human voice audios;

[0121] The human voice audio among the multiple human voice audios whose average loudness is greater than that of other human voice audios among the multiple human voice audios is determined as the human voice audio of the participant.

[0122] In one achievable manner, the determining module 950 is used to:

[0123] Among the at least one human voice audio, randomly play one human voice audio;

[0124] If the played human voice audio is the human voice audio of a non-participant, then in response to the audio change instruction triggered by the first account to the second account, the played human voice audio is replaced in the at least one human voice audio.

[0125] In one achievable manner, the determining module 950 is used to:

[0126] Determine whether there is a human voice audio in the at least one human voice audio whose voiceprint similarity with the participant audio in the participant audio library is greater than a similarity threshold, wherein the participant audio library stores the participant audio corresponding to each participant who joins the conference call;

[0127] If there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

[0128] In one achievable manner, the device further includes a shielding module, which is used to:

[0129] Determine and display the participant identification corresponding to the voice audio of the participant;

[0130] In response to a participant blocking instruction triggered by the first account and carrying a participant identifier corresponding to a designated participant, playing of the human voice audio corresponding to the designated participant is paused.

[0131] In an achievable manner, the conference call includes multiple conference topics, and the determination module 950 is used to:

[0132] Determine the conference topic currently being conducted by the conference call;

[0133] Among at least one human voice audio corresponding to each second account participating in the currently ongoing conference agenda, a target human voice audio to be played is determined.

[0134] Among them, the joining module 910, the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950 and the playing module 960 can all be implemented by software, or can be implemented by hardware. Exemplarily, the implementation of the determination module 950 is introduced by taking the determination module 950 as an example. Similarly, the implementation of the joining module 910, the acquisition module 920, the noise reduction module 930, the separation module 940 and the playing module 960 can refer to the implementation of the determination module 950.

[0135] As an example of a software functional unit, the module 950 can include code running on a computing instance. Among them, the computing instance can include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance can be one or more. For example, the module 950 can include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code can be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code can be distributed in the same availability zone (AZ) or in different AZs, each AZ including a data center or multiple data centers with close geographical locations. Among them, usually a region can include multiple AZs.

[0136] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0137] As an example of a hardware functional unit, the module 950 can include at least one computing device, such as a server, etc. Alternatively, the module A can also be a device implemented by an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD can be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

[0138] The multiple computing devices included in the determination module 950 may be distributed in the same region or in different regions. The multiple computing devices included in the determination module 950 may be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the determination module 950 may be distributed in the same VPC or in multiple VPCs. The multiple computing devices may be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0139] It should be noted that, in other embodiments, the determination module 950 can be used to execute any step in the method for playing call audio, and the joining module 910, the acquisition module 920, the noise reduction module 930, the separation module 940 and the playback module 960 can also be used to execute any step in the method for playing call audio. The steps that the joining module 910, the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950 and the playback module 960 are responsible for implementing can be specified as needed. By adding the module 910, the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950 and the playback module 960 respectively implement different steps in the method for playing call audio, the full functions of the device for playing call audio can be realized.

[0140] The present application also provides a computing device 100. Fig.10 As shown, the computing device 100 includes: a bus 102, a processor 104, a memory 106, and a communication interface 108. The processor 104, the memory 106, and the communication interface 108 communicate with each other through the bus 102. The computing device 100 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 100.

[0141] The bus 102 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 The bus 102 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 102 may include a path for transmitting information between various components of the computing device 100 (eg, the memory 106, the processor 104, and the communication interface 108).

[0142] The processor 104 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0143] The memory 106 may include a volatile memory, such as a random access memory (RAM). The memory 106 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0144] The memory 106 stores executable program codes, and the processor 104 executes the executable program codes to respectively implement the functions of the aforementioned joining module 910, the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950, and the playback module 960, thereby implementing the call audio playback method. That is, the memory 106 stores instructions for executing the call audio playback method.

[0145] The communication interface 108 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 100 and other devices or a communication network.

[0146] The embodiment of the present application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smart phone.

[0147] like Fig.11 As shown, the computing device cluster includes at least one computing device 100. The memory 106 in one or more computing devices 100 in the computing device cluster may store the same instructions for executing the method for playing call audio.

[0148] In some possible implementations, the memory 106 of one or more computing devices 100 in the computing device cluster may also store partial instructions for executing the method for playing the call audio. In other words, the combination of one or more computing devices 100 may jointly execute the instructions for executing the method for playing the call audio.

[0149] It should be noted that the memory 106 in different computing devices 100 in the computing device cluster can store different instructions, which are respectively used to execute part of the functions of the playback device of the call audio. That is, the instructions stored in the memory 106 in different computing devices 100 can implement the functions of one or more modules among the A module, the B module and the C module.

[0150] In some possible implementations, one or more computing devices in the computing device cluster may be connected via a network, which may be a wide area network or a local area network. Fig.12 A possible implementation is shown. Fig.12As shown, two computing devices 100A and 100B are connected via a network. Specifically, the network is connected via a communication interface in each computing device. In this type of possible implementation, the memory 106 in the computing device 100A stores instructions for executing the functions of the joining module 910. At the same time, the memory 106 in the computing device 100B stores instructions for executing the functions of the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950, and the playback module 960.

[0151] Fig.12 The connection method between the computing device clusters shown can be that considering that the call audio playback method provided in the present application requires a large amount of computing resources, it is considered that the functions implemented by the acquisition module 920, the noise reduction module 930, the separation module 940, the determination module 950 and the playback module 960 are handed over to the computing device 100B for execution.

[0152] It should be understood that Fig.12 The functions of the computing device 100A shown in FIG. 1 may also be completed by multiple computing devices 100. Similarly, the functions of the computing device 100B may also be completed by multiple computing devices 100.

[0153] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similar to that of Fig.11 and Fig.12 The connection mode of the computing device cluster is different in that the memory 106 in one or more computing devices 100 in the computing device cluster may store the same instructions for executing the method for playing the call audio.

[0154] In some possible implementations, the memory 106 of one or more computing devices 100 in the computing device cluster may also store partial instructions for executing the method for playing the call audio. In other words, the combination of one or more computing devices 100 may jointly execute the instructions for executing the method for playing the call audio.

[0155] The embodiment of the present application also provides a computer program product including instructions. The computer program product may be software or a program product including instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes a method for playing call audio.

[0156] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by the computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute a method for playing call audio.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for playing call audio, characterized in that: The method comprises: In response to a conference call joining operation of the first account, joining the first account to the conference call; Acquire audio data of multiple second accounts in the conference call, wherein the audio data of at least one second account includes multiple human voice audios; Performing noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account; Performing human voice separation on the noise reduction processed audio data corresponding to each second account to obtain at least one human voice audio corresponding to each second account; Determining a target human voice audio to be played from the at least one human voice audio corresponding to each second account; The target human voice audio corresponding to each second account is played.

2. The method according to claim 1, characterized in that The performing noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account includes: For each audio data of the second account, extracting audio features corresponding to the audio data; Inputting the audio feature into a first speech detection algorithm to obtain a first probability that each audio frame in the audio data belongs to human voice audio; Inputting the audio feature into a second speech detection algorithm to obtain a second probability that each audio frame in the audio data belongs to human voice audio; Merging the first probability and the second probability that each audio frame belongs to human voice audio to obtain a third probability that each audio frame belongs to human voice audio; Based on the third probability that each audio frame belongs to human voice audio, the audio data is subjected to noise reduction processing to obtain audio data after noise reduction processing.

3. The method according to claim 1 or 2, characterized in that: The step of determining a target human voice audio to be played from the at least one human voice audio corresponding to each second account includes: For at least one human voice audio corresponding to each second account, determine the human voice audio of the participant in the at least one human voice audio; The human voice audio of the participant is determined as the target human voice audio to be played.

4. The method according to claim 3, characterized in that The step of determining the human voice audio of the participant in the at least one human voice audio comprises: In a case where the number of the at least one human voice audio is plural, determining average loudnesses respectively corresponding to the plurality of human voice audios; The human voice audio among the multiple human voice audios whose average loudness is greater than that of other human voice audios among the multiple human voice audios is determined as the human voice audio of the participant.

5. The method according to claim 3, characterized in that: The step of determining the human voice audio of the participant in the at least one human voice audio comprises: Among the at least one human voice audio, randomly play one human voice audio; If the played human voice audio is the human voice audio of a non-participant, then in response to the audio change instruction triggered by the first account to the second account, the played human voice audio is replaced in the at least one human voice audio.

6. The method according to claim 3, characterized in that The step of determining the human voice audio of the participant in the at least one human voice audio comprises: Determine whether there is a human voice audio in the at least one human voice audio whose voiceprint similarity with the participant audio in the participant audio library is greater than a similarity threshold, wherein the participant audio library stores the participant audio corresponding to each participant who joins the conference call; If there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

7. The method according to claim 6, characterized in that The method further comprises: Determine and display the participant identification corresponding to the voice audio of the participant; In response to a participant blocking instruction triggered by the first account and carrying a participant identifier corresponding to a designated participant, playing of the human voice audio corresponding to the designated participant is paused.

8. The method according to any one of claims 1 to 7, characterized in that: The conference call includes a plurality of conference topics, and determining a target human voice audio to be played from the at least one human voice audio corresponding to each second account includes: Determine the conference topic currently being conducted by the conference call; Among at least one human voice audio corresponding to each second account participating in the currently ongoing conference agenda, a target human voice audio to be played is determined.

9. A device for playing call audio, characterized in that: The device comprises: A joining module, configured to, in response to a conference call joining operation of a first account, join the first account to the conference call; An acquisition module, configured to acquire audio data of a plurality of second accounts in the conference call, wherein the audio data of at least one second account includes a plurality of human voice audios; A noise reduction module, used to perform noise reduction processing on the audio data of each second account to obtain the noise-reduced audio data corresponding to each second account; A separation module, configured to perform human voice separation on the noise reduction processed audio data corresponding to each second account, to obtain at least one human voice audio corresponding to each second account; a determination module, configured to determine a target human voice audio to be played from the at least one human voice audio corresponding to each second account; The playing module is used to play the target human voice audio corresponding to each second account.

10. The device according to claim 9, characterized in that The noise reduction module is used to: For each audio data of the second account, extracting audio features corresponding to the audio data; Inputting the audio feature into a first speech detection algorithm to obtain a first probability that each audio frame in the audio data belongs to human voice audio; Inputting the audio feature into a second speech detection algorithm to obtain a second probability that each audio frame in the audio data belongs to human voice audio; Merging the first probability and the second probability that each audio frame belongs to human voice audio to obtain a third probability that each audio frame belongs to human voice audio; Based on the third probability that each audio frame belongs to human voice audio, the audio data is subjected to noise reduction processing to obtain audio data after noise reduction processing.

11. The device according to claim 9 or 10, characterized in that The determining module is used to: For at least one human voice audio corresponding to each second account, determine the human voice audio of the participant in the at least one human voice audio; The human voice audio of the participant is determined as the target human voice audio to be played.

12. The device according to claim 11, characterized in that The determining module is used to: In a case where the number of the at least one human voice audio is plural, determining average loudnesses respectively corresponding to the plurality of human voice audios; The human voice audio among the multiple human voice audios whose average loudness is greater than that of other human voice audios among the multiple human voice audios is determined as the human voice audio of the participant.

13. The device according to claim 11, characterized in that The determining module is used to: Among the at least one human voice audio, randomly play one human voice audio; If the played human voice audio is the human voice audio of a non-participant, then in response to the audio change instruction triggered by the first account to the second account, the played human voice audio is replaced in the at least one human voice audio.

14. The device according to claim 11, characterized in that The determining module is used to: Determine whether there is a human voice audio in the at least one human voice audio whose voiceprint similarity with the participant audio in the participant audio library is greater than a similarity threshold, wherein the participant audio library stores the participant audio corresponding to each participant who joins the conference call; If there is human voice audio greater than the similarity threshold, the human voice audio greater than the similarity threshold is determined as the human voice audio of the participant.

15. The device according to claim 14, characterized in that The device also includes a shielding module, which is used for: Determine and display the participant identification corresponding to the voice audio of the participant; In response to a participant blocking instruction triggered by the first account and carrying a participant identifier corresponding to a designated participant, playing of the human voice audio corresponding to the designated participant is paused.

16. The device according to any one of claims 9 to 15, characterized in that The conference call includes multiple conference topics, and the determination module is used to: Determine the conference topic currently being conducted by the conference call; Among at least one human voice audio corresponding to each second account participating in the currently ongoing conference agenda, a target human voice audio to be played is determined.

17. A computing device cluster, characterized in that: comprising at least one computing device, each computing device comprising a processor and a memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster executes the method according to claims 1 to 8.

18. A computer program product comprising instructions, characterized in that When the instructions are executed by a computing device, the computing device is caused to perform the method according to claims 1 to 8.

19. A computer-readable storage medium, characterized in that: The method comprises computer program instructions which, when executed by a computing device, cause the computing device to perform the method of claims 1 to 8.

Images