CN115440229A - Audio processing method, device, processor and system - Google Patents

Abstract

The present application provides an audio processing method, device, processor and system, the method comprising: acquiring at least one audio clip, and using a voiceprint recognition model to perform voiceprint recognition on at least one audio clip to obtain a first recognition result; When the first recognition result indicates that at least one audio segment is a non-target mute segment and the duration of at least one audio segment is greater than or equal to the first duration threshold, obtain the highest recognition score in the first recognition result; in the audio of at least one audio segment When the duration is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold, it is determined that the character corresponding to at least one audio clip is an unknown character; and the unknown character is registered in the voiceprint recognition model library. The method achieves the technical effect of voice role separation through an unknown role separation algorithm, and solves the technical problem that the speaker's voiceprint usually needs to be registered in advance when performing role separation.

Description

Audio processing method, device, processor and system

technical field

The present application relates to the field of data processing, in particular, to an audio processing method, device, processor and system.

Background technique

At present, both the audio conferencing system and the recorded interrogation system need to use a role separation system, which is used to separate the voices of multiple speakers, and perform voice transcription or speaker role display based on the separation results.

However, the current role separation technology usually needs to register the speaker's voiceprint in advance when performing voiceprint role separation. In actual application scenarios, its ease of use is poor and the cost of preparation is high.

The above information disclosed in the Background section is only to enhance the understanding of the background of the technology described herein, therefore, the Background may contain certain information which is not formed in the country for those skilled in the art. known prior art.

Contents of the invention

The main purpose of the present application is to provide an audio processing method, device, processor and system to solve the problem in the prior art that the speaker's voiceprint needs to be registered in advance when performing voiceprint role separation.

According to an aspect of an embodiment of the present invention, an audio processing method is provided, including: acquiring at least one audio clip, and performing voiceprint recognition on the at least one audio clip using a voiceprint recognition model to obtain a first recognition result; When the first recognition result indicates that the at least one audio segment is a non-target mute segment and the duration of the at least one audio segment is greater than or equal to a first duration threshold, obtain the highest recognition score in the first recognition result ; When the audio duration of the at least one audio clip is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold, determine that the role corresponding to the at least one audio clip is an unknown character, and the second duration The threshold is greater than the first duration threshold; and the unknown character is registered in the voiceprint recognition model library.

Optionally, when the audio duration of the at least one audio segment is greater than or equal to a second duration threshold and the highest recognition score is less than the score threshold, determining that the character corresponding to the at least one audio segment is an unknown character includes : a first determining step, when the audio duration of the at least one audio segment is less than the second duration threshold and the highest recognition score is less than the score threshold, determine that the role corresponding to the at least one audio segment is a candidate Unknown role; the second determining step is to obtain the subsequent audio segment of the at least one audio segment to obtain a first updated audio segment, and perform the voiceprint recognition on the first updated audio segment to obtain a second recognition result, and When the second recognition result indicates that the audio duration of the first updated audio clip is greater than or equal to the second duration threshold and the highest recognition score is greater than the score threshold, the candidate unknown character is updated as already If the second recognition result indicates that the audio duration of the first updated audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than or equal to the score threshold, the candidate The unknown character is updated to the unknown character; when the second recognition result indicates that the audio duration of the first updated audio clip is less than the second duration threshold, the second determining step is repeatedly executed until it is determined that the The character corresponding to the first updated audio clip is the known character or the unknown character.

Optionally, when the audio duration of the at least one audio segment is less than the second duration threshold and the highest recognition score is less than the score threshold, determine that the character corresponding to the at least one audio segment is a candidate unknown character , including: when the audio duration of the at least one audio segment is less than a first duration threshold and the highest recognition score is less than a first score threshold, determining that the character corresponding to the at least one audio segment is the candidate unknown character; When the audio duration of the at least one audio segment is greater than or equal to the first duration threshold and less than a third duration threshold and the highest recognition score is greater than or equal to the first score threshold and less than a second score threshold, determining that the character corresponding to the at least one audio clip is the candidate unknown character, the first duration threshold is less than the third duration threshold, and the first score threshold is less than the second score threshold; in the at least one When the audio duration of the audio clip is greater than or equal to the third duration threshold and less than the second duration threshold and the highest recognition score is greater than or equal to the second score threshold and less than the third score threshold, determine the at least one The character corresponding to the audio clip is the candidate unknown character, the third duration threshold is smaller than the second duration threshold, and the second score threshold is smaller than the third score threshold.

Optionally, when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is smaller than the score threshold, it is determined that the role corresponding to the at least one audio segment is unknown The role includes: when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is smaller than the score threshold, determining that the role corresponding to the at least one audio segment is the The unknown role; when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is greater than or equal to the score threshold, determine the role corresponding to the at least one audio segment is a known role.

Optionally, acquiring at least one audio clip, and using a voiceprint recognition model to perform voiceprint recognition on the at least one audio clip to obtain the first recognition result includes: a third determining step, where the first recognition result In the case of characterizing that the at least one audio segment is the target silence segment, acquiring the duration of the at least one audio segment, and determining that the at least one audio segment is longer than a fourth duration threshold when the duration of the at least one audio segment is greater than a fourth duration threshold The segment is the target mute segment, and the corresponding role is empty; the fourth determining step is to obtain the at least one audio segment when the duration of the at least one audio segment is less than or equal to the fourth duration threshold Subsequent audio clips, obtain a second updated audio clip, and perform the voiceprint recognition on the second updated audio clip to obtain a third recognition result, where the third recognition result indicates that the duration of the second updated audio is longer than the specified In the case of the fourth duration threshold, determine that the at least one audio segment is the target silent segment; when the third recognition result indicates that the audio duration of the second updated audio segment is less than or equal to the fourth duration threshold In some cases, the fourth determining step is repeatedly executed until it is determined that the second updated audio segment is the target silent segment or the non-target silent segment.

Optionally, when the duration of the at least one audio segment is greater than the fourth duration threshold, determining the at least one audio segment as the target mute segment includes: when the duration of the at least one audio segment is greater than In the case of the second duration threshold, determine that the at least one audio segment is the target mute segment; in the case where the duration of the at least one audio segment is less than or equal to the second duration threshold and greater than the fourth duration threshold Next, it is determined that the at least one audio segment is the target silence segment, and the second duration threshold is greater than the fourth duration threshold.

Optionally, the method further includes: a fifth determining step, if the historical role is not empty, determine whether the historical role is the same as the current role, wherein the current role corresponds to the current at least one audio segment The role of the historical role is the role corresponding to the audio clip before the at least one audio clip; the sixth determination step, in the case that the historical role is the same as the current role, it is determined that no role switching has occurred; the seventh The determining step is to determine whether the duration of the at least one audio segment is greater than or equal to a second duration threshold if the historical role is different from the current role, and if the duration of the at least one audio segment is greater than or equal to the In the case of the second duration threshold, it is determined that the role switching occurs; in the case that the duration of the at least one audio segment is less than the second duration threshold, the subsequent audio segment of the at least one audio segment is acquired, and the second audio segment is obtained. 3. Update the audio clip, and repeat the fifth determination step to the seventh determination step at least once until it is determined that the role switching occurs or the role switching does not occur. During the repeated execution, the current role The role corresponding to the third update audio segment.

According to another aspect of the embodiments of the present invention, there is provided an audio processing device, the processing device includes: a first acquisition unit, configured to acquire at least one audio segment, and use a voiceprint recognition model to identify the at least one audio segment Perform voiceprint recognition to obtain a first recognition result; a second acquisition unit, configured to indicate that the at least one audio segment is a non-target silent segment and the duration of the at least one audio segment is greater than or equal to the first recognition result In the case of a duration threshold, the highest recognition score in the first recognition result is obtained; the determination unit is configured to, when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than the score In the case of a threshold, it is determined that the role corresponding to the at least one audio clip is an unknown character, and the second duration threshold is greater than the first duration threshold; a registration unit is configured to register the unknown character to the voiceprint recognition in the model library.

According to still another aspect of the present application, a processor is provided, and the processor is used to run a program, wherein any one of the processing methods is executed when the program is running.

According to still another aspect of the present application, an audio processing system is provided, and the processing system includes: a voiceprint recognition system, one or more processors, memory and one or more programs, wherein the one or more A program is stored in the memory and is configured to be executed by the one or more processors, the one or more programs including for performing any one of the processing methods described.

In the embodiment of the present invention, the method of voiceprint recognition is adopted, and through the unknown role separation algorithm, the purpose of automatically identifying multiple unknown roles in the audio is achieved, thereby realizing the technical effect of voice role separation, and further solving the problem of character identification. Separation usually requires the technical issue of registering the speaker's voiceprint in advance.

Description of drawings

The accompanying drawings constituting a part of the present application are used to provide further understanding of the present application, and the schematic embodiments and descriptions of the present application are used to explain the present application, and do not constitute improper limitations to the present application. In the attached picture:

Fig. 1 shows a schematic flow chart of an embodiment of an audio processing method according to the present application;

Fig. 2 shows a schematic flow chart of unknown character detection according to yet another embodiment of an audio processing method of the present application;

FIG. 3 shows a schematic flow diagram of a silence detection process according to yet another embodiment of an audio processing method of the present application;

FIG. 4 shows a schematic diagram of a role switching process according to another embodiment of an audio processing method of the present application;

Fig. 5 shows a schematic diagram of an embodiment of an audio processing device according to the present application.

detailed description

It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiment of the application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the application. Obviously, the described embodiment is only It is an embodiment of a part of the application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

It should be noted that the terms "first" and "second" in the description and claims of the present application and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It should be understood that the data so used may be interchanged under appropriate circumstances for the embodiments of the application described herein. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

For ease of description, some nouns or terms involved in the embodiments of the present application are described below:

Voiceprint recognition: a technology that extracts the speaker's voice characteristics and speech content information, and automatically verifies the speaker's identity;

Automatic Speech Recognition Technology: A technology that converts human speech into text.

As mentioned in the background technology, when performing voiceprint role separation in the prior art, the problem of voiceprint role separation can only be performed by registering the speaker’s voiceprint in advance. In order to solve the above problems, a typical implementation of this application In the manner, an audio processing method, device, processor and system are provided.

Fig. 1 is a flowchart of an audio processing method according to an embodiment of the present application. As shown in Figure 1, the method includes the following steps:

Step S101, acquire at least one audio clip, and use a voiceprint recognition model to perform voiceprint recognition on the at least one audio clip to obtain a first recognition result.

In the process of obtaining audio clips, it can be obtained by any one or more methods of ordinary microphone audio collection, array microphone audio collection, hand-in-hand microphone audio collection, computer speakers, mobile phone recording, and network audio collection. Mechanisms are diverse and can be applied to different scenarios.

The aforementioned at least one audio segment may be one audio segment, or may be multiple audio segments. In different application scenarios, the number of audio segments may be different.

The voiceprint recognition model of the present application can be any feasible voiceprint recognition model in the prior art, specifically it can be a template model, and it can also be a random model. The characteristic parameters are compared, and the distortion between the two is used as the similarity. For example, the VQ (vector quantization) model is the vector quantization model and the DTW (dynamic time warping) model is the dynamic time warping model; where the VQ model is clustered and quantized. The method generates a codebook, quantizes and encodes the test data during recognition, and uses the degree of distortion as the judgment standard. The DTW model compares the input feature vector sequence to be recognized with the feature vector extracted during training, and matches the optimal path. The random model is a parameter model, which uses a probability density function to simulate the speaker. The training process is used to predict the parameters of the probability density function. The matching process is completed by calculating the similarity of the test sentences of the corresponding model, such as GMM The model is the Gaussian mixture model, which is one of the best and most commonly used models in text-independent speaker recognition. The HMM model, the hidden Markov model, is used to describe a Markov process with hidden unknown parameters. statistical model. More specifically, the voiceprint recognition model includes a training phase and a testing phase, wherein the training phase includes four parts: training voice, feature extraction, model training, and model library; the testing phase includes testing voice, feature extraction, and scoring judgment.

Step S102, when the first recognition result indicates that at least one audio segment is a non-target silent segment and the duration of at least one audio segment is greater than or equal to the first duration threshold, obtaining the highest recognition score in the first recognition result;

The above-mentioned first recognition result at least includes information indicating whether at least one audio segment is a non-target silent segment, duration information of at least one audio segment, and a corresponding recognition score of at least one audio segment. In the above steps, if at least one audio segment is a non-target silent segment, it means that the at least one audio segment is not a silent segment, that is, the highest recognition score is obtained only when at least one audio segment is not a silent segment, because , when at least one audio segment is a silent segment, the at least one silent segment does not correspond to any character, so identification of an unknown character is not involved. In addition, if the duration of at least one audio segment is too short and less than the first duration threshold, the determined character may be inaccurate. Therefore, another prerequisite for obtaining the highest recognition score is that the duration of at least one audio segment is greater than or equal to the first duration threshold. duration threshold.

In addition, the above-mentioned highest recognition score refers to the highest score obtained after at least one audio segment is matched with a role in the model library in the voiceprint recognition model.

Step S103, when the audio duration of at least one audio clip is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold, determine that the character corresponding to the at least one audio clip is an unknown character, and the second duration threshold is greater than the first duration threshold ;

In the above steps, if the audio clip is too short, it cannot be accurately determined whether the corresponding character is an unknown character. Therefore, the audio duration of at least one audio clip is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold. In this case, it is considered that there is no character matching the audio in the model library, that is, the character corresponding to the audio is considered to be an unknown character.

Step S104, registering the unknown character into the library of the voiceprint recognition model.

In the above audio processing method, by comparing the duration of the audio clip and the relationship between the highest recognition score and the corresponding threshold, it can be determined whether the character corresponding to the audio is an unknown character, and if it is determined to be an unknown character, register the unknown character in the voiceprint In the identified model library, voiceprint role separation can be performed later without pre-registration, thereby solving the problem in the prior art that requires pre-registration to perform voiceprint role separation. Compared with the prior art, this solution requires pre-registration For the solution, it is easy to use and applicable to a wide range of scenarios.

In a specific embodiment of the present application, on the basis of including the above-mentioned steps S101 to S104, the above-mentioned step S103 is further refined, and this step specifically includes: Step S1031, the first determining step, in the above-mentioned at least one When the audio duration of the segment is less than the above-mentioned duration threshold and the above-mentioned highest recognition score is less than the above-mentioned score threshold, determine that the character corresponding to the at least one audio segment is a candidate unknown character, that is, first determine at least one audio segment that may be an unknown character; step S1032 , the second determining step is to obtain the subsequent audio segment of the at least one audio segment, obtain the first updated audio segment, and perform the above-mentioned voiceprint recognition on the first updated audio segment to obtain a second recognition result, in the above-mentioned second recognition result When the audio duration representing the first updated audio segment is greater than or equal to the duration threshold and the highest recognition score is greater than the score threshold, update the candidate unknown character to a known character, and when the second recognition result represents the first When the audio duration of the updated audio clip is greater than or equal to the above-mentioned duration threshold and the above-mentioned highest recognition score is less than or equal to the above-mentioned score threshold, update the above-mentioned candidate unknown character to the above-mentioned unknown character, that is, after determining the possible unknown character, obtain subsequent Audio clips, identifying subsequent audio clips, and determining whether the candidate unknown character is an unknown character according to the recognition results of the subsequent audio clips; step S1033, when the second recognition result indicates that the audio duration of the first updated audio clip is less than the above duration threshold In some cases, the above second determining step is repeatedly executed until it is determined that the character corresponding to the above first updated audio segment is the above known character or the above unknown character. In this method, the candidate unknown character is determined first, and then the subsequent audio clip is obtained for determination, so that the result of determining whether the character is an unknown character is more accurate.

In one embodiment of the present application, on the basis of including the above-mentioned steps S1031 to S1033, the specific above-mentioned step S1031 is further refined. FIG. 2 is a flow chart of unknown character detection according to the embodiment of the present application, as shown in FIG. 2 The method includes the following steps: when the audio duration of the at least one audio segment is less than the duration threshold and the highest recognition score is less than the score threshold, determining that the character corresponding to the at least one audio segment is a candidate unknown character includes: When the audio duration of the at least one audio segment is less than the first duration threshold and the highest recognition score is less than the first score threshold, it is determined that the character corresponding to the at least one audio segment is the candidate unknown character; When the duration is greater than or equal to the first duration threshold and less than the third duration threshold and the highest recognition score is greater than or equal to the first score threshold and less than the second score threshold, determine that the role corresponding to the at least one audio segment is the candidate Unknown character, the first duration threshold is less than the third duration threshold, the first score threshold is less than the second score threshold; the audio duration of the at least one audio segment is greater than or equal to the third duration threshold and less than the second duration threshold And when the above-mentioned highest recognition score is greater than or equal to the above-mentioned second score threshold and smaller than the third score threshold, it is determined that the character corresponding to the above-mentioned at least one audio segment is the above-mentioned candidate unknown character, and the above-mentioned third duration threshold is less than the above-mentioned second duration threshold, The second score threshold is smaller than the third score threshold.

In the above steps, in order to improve the sensitivity of model identification, a variety of different duration thresholds and score thresholds are set. This application sets three duration thresholds and three score thresholds. In the above steps, by comparing the duration of the audio clip and the relationship between the highest recognition score and different corresponding thresholds, the candidate unknown character, that is, the possible unknown character, can be determined first, and then wait for subsequent determination, which can improve the accuracy of the recognition model. Therefore, there are three situations in which the character corresponding to the audio can be considered as a candidate unknown character, that is, a possible unknown character. In the first case: the audio duration of at least one audio segment is less than the first duration threshold and the above-mentioned highest recognition score is less than the first score threshold; The second case: the audio duration of at least one audio clip is greater than or equal to the above-mentioned first duration threshold and less than the third duration threshold and the highest recognition score is greater than or equal to the above-mentioned first score threshold and less than the second score threshold; the third case: The audio duration of at least one audio segment is greater than or equal to the third duration threshold and less than the second duration threshold and the highest recognition score is greater than or equal to the second score threshold and less than the third score threshold. In an embodiment of the present application, on the basis of including the above-mentioned steps S101 to S104, the above-mentioned step S103 is further refined, and this step specifically includes: when the audio duration of the above-mentioned at least one audio segment is greater than or equal to the above-mentioned second When the duration threshold and the above-mentioned highest recognition score are less than the above-mentioned score threshold, it is determined that the character corresponding to the above-mentioned at least one audio segment is the above-mentioned unknown character, that is, the audio duration of at least one audio segment is greater than or equal to the above-mentioned second duration threshold and the above-mentioned highest When the recognition score is less than the aforementioned score threshold, it may be determined that the character corresponding to at least one audio segment is an unknown character; when the audio duration of the at least one audio segment is greater than or equal to the aforementioned second duration threshold and the aforementioned highest recognition score is greater than or equal to the aforementioned score In the case of threshold value, it is determined that the character corresponding to the at least one audio segment is a known character, that is, when the audio duration of at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is greater than the score threshold, It may be determined that the character corresponding to the at least one audio segment is a known character.

Through the above steps, it can be determined whether the character corresponding to at least one audio clip is known or unknown, and the recognition of the unknown character is realized. After it is determined that the character corresponding to the audio is a known character or an unknown character, the automatic speech recognition technology can also be called to convert the speech into text or call the machine translation technology. In addition, it can also display the audio role name and the result of audio conversion to text, and then realize the functions of audio and processing result comparison and playback, audio clip selection, and audio role name modification.

In the actual voiceprint recognition process, the audio can be muted. The purpose is to identify the silent segment in the audio. Only when the audio is not silent can the subsequent role separation and role switching recognition be continued, so as to improve the efficiency of audio recognition. effect. In an embodiment of the present application, on the basis of including the above steps S101 to S104, the above step S101 is refined. FIG. 3 is a flow chart of silence detection according to the embodiment of the present application. As shown in Figure 3, the method includes the following steps: step S1011, the third determining step, when the above-mentioned first recognition result indicates that the above-mentioned at least one audio segment is the above-mentioned target mute segment, obtain the duration of the above-mentioned at least one audio segment, and When the duration of the at least one audio segment is greater than the fourth duration threshold, it is determined that the at least one audio segment is the target mute segment, and the corresponding role is empty, that is, at least one audio segment is considered to be mute and there is no corresponding role; step S1012, the fourth determination step, in the case that the duration of the at least one audio segment is less than or equal to the fourth duration threshold, acquire the subsequent audio segment of the at least one audio segment, obtain the second updated audio segment, and perform the above-mentioned second Update the audio segment to perform the above-mentioned voiceprint recognition to obtain a third recognition result, and if the above-mentioned third recognition result indicates that the duration of the updated audio is greater than the fourth duration threshold, determine that the at least one audio segment is the target silent segment, that is, determine After the mute segment is obtained, the subsequent audio segment is obtained, and the subsequent audio segment is identified, and the mute segment or the non-mute segment is determined according to the recognition result of the subsequent audio segment; Step S1013, the above-mentioned second update audio is represented by the above-mentioned third recognition result If the audio duration of the segment is less than or equal to the fourth duration threshold, the fourth determination step is repeatedly executed until it is determined that the second updated audio segment is the target silent segment or the non-target silent segment. In this method, the target silent segment is determined first, and then the subsequent audio segment is obtained for determination, so that the result of determining whether the target silent segment is obtained is more accurate.

In an embodiment of the present application, on the basis of including the above-mentioned steps S1011 to S1013, the specific above-mentioned step S1011 is refined, and this step specifically includes: when the duration of the above-mentioned at least one audio segment is greater than the above-mentioned fourth duration threshold Determining the at least one audio segment as the target mute segment includes: determining the at least one audio segment as the target mute segment when the duration of the at least one audio segment is greater than a third duration threshold, that is, determining the target mute segment Finally, if the duration is greater than the third duration threshold, it can be determined as the target silence segment; if the duration of the at least one audio segment is less than or equal to the third duration threshold and greater than the fourth duration threshold, determine the at least one audio segment. An audio segment is the above-mentioned target mute segment, and the above-mentioned third duration threshold is greater than the above-mentioned fourth duration threshold, that is, after the target mute segment is determined, if the duration is less than or equal to the above-mentioned third duration threshold and greater than the above-mentioned fourth duration threshold, Can be identified as the target silent segment. Through the above steps, the setting of the fourth duration threshold can eliminate the silence in which the speaker pauses in the audio, and can make the voiceprint recognition result more accurate.

In the actual use process, there are still cases where the speaker changes, so it is also necessary to perform role switching detection on the audio. In another embodiment of the present application, on the basis of including the above steps S101 to S104, FIG. 4 is a flow chart of role switching according to the embodiment of the present application. As shown in FIG. 4, the method includes the following steps: fifth In the determining step, if the historical role is not empty, determine whether the historical role is the same as the current role, wherein the current role is the role corresponding to the current at least one audio segment, and the historical role is the one before the at least one audio segment The role corresponding to the audio clip, that is, first determine whether the role corresponding to at least one audio clip is the same as that of the previous audio clip; the sixth determination step, in the case that the above-mentioned historical role is the same as the above-mentioned current role, determine that no role switching occurs, that is, satisfy In the case where at least one audio clip corresponds to the same role as the previous audio clip, there is no role switching; the seventh determining step is to determine the duration of the at least one audio clip when the above-mentioned historical role is different from the above-mentioned current role Whether it is greater than or equal to the third duration threshold, in the case that the duration of the at least one audio segment is greater than or equal to the third duration threshold, it is determined that the role switching occurs, that is, it is determined that the role corresponding to at least one audio segment is different from the previous audio segment In the same case, if the duration is greater than or equal to the third duration threshold, it is determined that a role switch has occurred; when the duration of the at least one audio segment is less than the third duration threshold, obtain the at least one audio segment. Subsequent audio clips, to obtain the third updated audio clip, repeat the above-mentioned fifth determination step to the above-mentioned seventh determination step at least once until it is determined that the above-mentioned role switching occurs or the above-mentioned role switching does not occur. During the repeated execution process, the above-mentioned current The role is the role corresponding to the above-mentioned third updated audio segment, that is, after determining that the role switch occurs, the subsequent audio segment is obtained, the subsequent audio segment is identified, and whether the role switch occurs is determined according to the identification result of the subsequent audio segment. In this method, by first determining whether the role switching occurs, and then determining whether the role switching occurs after acquiring subsequent audio clips, the result of determining whether the role switching occurs is more accurate. Through this method, it is possible to accurately identify whether a role switch occurs in the audio.

It should be noted that the steps shown in the flowcharts of the accompanying drawings may be performed in a computer system, such as a set of computer-executable instructions, and that although a logical order is shown in the flowcharts, in some cases, The steps shown or described may be performed in an order different than here.

The embodiment of the present application also provides an audio processing device. It should be noted that the audio processing device in the embodiment of the present application can be used to implement the audio processing method provided in the embodiment of the present application. The audio processing device provided by the embodiment of the present application is introduced below.

Fig. 5 is a schematic diagram of an audio processing device according to an embodiment of the present application. As shown in Figure 5, the device includes:

The first obtaining unit 10 is configured to obtain at least one audio segment, and perform voiceprint recognition on the at least one audio segment by using a voiceprint recognition model to obtain a first recognition result.

In the process of obtaining audio clips, it can be obtained by any one or more methods of ordinary microphone audio collection, array microphone audio collection, hand-in-hand microphone audio collection, computer speakers, mobile phone recording, and network audio collection. Mechanisms are diverse and can be applied to different scenarios.

The aforementioned at least one audio segment may be one audio segment, or may be multiple audio segments. In different application scenarios, the number of audio segments may be different.

The voiceprint recognition model of the present application can be any feasible voiceprint recognition model in the prior art, specifically it can be a template model, and it can also be a random model. The characteristic parameters are compared, and the distortion between the two is used as the similarity. For example, the VQ (vector quantization) model is the vector quantization model and the DTW (dynamic time warping) model is the dynamic time warping model; where the VQ model is clustered and quantized. The device generates a codebook, quantizes and encodes the test data during recognition, and uses the degree of distortion as the criterion for judgment. The DTW model compares the input feature vector sequence to be recognized with the feature vector extracted during training, and uses the optimal path matching The random model is a parameter model, which uses a probability density function to simulate the speaker. The training process is used to predict the parameters of the probability density function. The matching process is completed by calculating the similarity of the test sentences of the corresponding model, such as GMM The model is the Gaussian mixture model, which is one of the best and most commonly used models in text-independent speaker recognition. The HMM model, the hidden Markov model, is used to describe a Markov process with hidden unknown parameters. statistical model. More specifically, the voiceprint recognition model includes a training phase and a testing phase, wherein the training phase includes four parts: training voice, feature extraction, model training, and model library; the testing phase includes testing voice, feature extraction, and scoring judgment.

The second obtaining unit 20 is configured to obtain the highest value in the first recognition result when the first recognition result indicates that at least one audio segment is a non-target silent segment and the duration of the at least one audio segment is greater than or equal to the first duration threshold recognition score;

The above-mentioned first recognition result at least includes information indicating whether at least one audio segment is a non-target silent segment, duration information of at least one audio segment, and a corresponding recognition score of at least one audio segment. In the above unit, if at least one audio segment is a non-target silent segment, it means that the at least one audio segment is not a silent segment, that is, the highest recognition score is obtained only when at least one audio segment is not a silent segment, because , when at least one audio segment is a silent segment, the at least one silent segment does not correspond to any character, so identification of an unknown character is not involved. In addition, if the duration of at least one audio segment is too short and less than the first duration threshold, the determined character may be inaccurate. Therefore, another prerequisite for obtaining the highest recognition score is that the duration of at least one audio segment is greater than or equal to the first duration threshold. duration threshold.

In addition, the above-mentioned highest recognition score refers to the highest score obtained after at least one audio segment is matched with a role in the model library in the voiceprint recognition model.

The determining unit 30 is configured to determine that the character corresponding to the at least one audio segment is an unknown character when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold. The second duration threshold is greater than the first duration threshold;

In the above unit, if the audio clip is too short, it cannot be accurately determined whether the corresponding character is an unknown character. Therefore, the audio duration of at least one audio clip is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold. In this case, it is considered that there is no character matching the audio in the model library, that is, the character corresponding to the audio is considered to be an unknown character.

The registration unit 40 is configured to register the above-mentioned unknown character into the above-mentioned voiceprint recognition model library.

In the above audio processing device, by comparing the duration of the audio segment and the relationship between the highest recognition score and the corresponding threshold, it can be determined whether the character corresponding to the audio is an unknown character, and if it is determined to be an unknown character, register the unknown character in the voiceprint In the identified model library, voiceprint role separation can be performed later without pre-registration, thereby solving the problem in the prior art that requires pre-registration to perform voiceprint role separation. Compared with the prior art, this solution requires pre-registration For the solution, it is easy to use and applicable to a wide range of scenarios.

In a specific embodiment of the present application, on the basis of including the first acquisition unit, the second acquisition unit, the determination unit and the registration unit, the specific determination unit is further refined, and the determination unit includes a first determination module , a second determination module and a third determination module, wherein the first determination module is configured to determine the at least one audio segment when the audio duration of the at least one audio segment is less than the duration threshold and the highest recognition score is less than the score threshold The corresponding role is a candidate unknown character, that is, first determine at least one audio clip that may be an unknown character; the second determination module is used to obtain the subsequent audio clip of the at least one audio clip, obtain the first updated audio clip, and perform the above-mentioned first Updating the audio segment to perform the above-mentioned voiceprint recognition to obtain a second recognition result, when the second recognition result indicates that the audio duration of the first updated audio segment is greater than or equal to the above-mentioned duration threshold and the above-mentioned highest recognition score is greater than the above-mentioned score threshold, Updating the above-mentioned candidate unknown character to a known character, when the above-mentioned second recognition result indicates that the audio duration of the above-mentioned first updated audio segment is greater than or equal to the above-mentioned duration threshold and the above-mentioned highest recognition score is less than or equal to the above-mentioned score threshold, the above-mentioned candidate The unknown character is updated to the above unknown character, that is, after the possible unknown character is determined, the subsequent audio clip is obtained, the subsequent audio clip is identified, and whether the candidate unknown character is an unknown character is determined according to the recognition result of the subsequent audio clip; the third The determination module is used to repeatedly execute the second determination module until it is determined that the role corresponding to the first updated audio segment is the above-mentioned Known roles or the aforementioned unknown roles. In the device, by first determining the candidate unknown character, and then obtaining subsequent audio clips for determination, the result of determining whether the unknown character is determined is more accurate.

In an embodiment of the present application, on the basis of including the first determination module, the second determination module and the third determination module, the above-mentioned first determination module is further refined, and the module specifically includes the first determination module module, a second determination sub-module and a third determination sub-module, wherein the first determination sub-module is used for when the audio duration of the at least one audio segment is less than the first duration threshold and the highest recognition score is less than the first score threshold, Determining that the character corresponding to the at least one audio segment is the candidate unknown character; the second determining submodule is used to determine when the audio duration of the at least one audio segment is greater than or equal to the first duration threshold and less than the third duration threshold and the highest recognition score When it is greater than or equal to the first score threshold and less than the second score threshold, it is determined that the character corresponding to the at least one audio segment is the candidate unknown character, the first duration threshold is less than the third duration threshold, and the first score threshold Less than the above-mentioned second score threshold; the third determination submodule is used to determine that the audio duration of the at least one audio segment is greater than or equal to the above-mentioned third duration threshold and less than the second duration threshold and the above-mentioned highest recognition score is greater than or equal to the above-mentioned second score threshold and is less than the third score threshold, determine that the character corresponding to the at least one audio segment is the candidate unknown character, the third duration threshold is smaller than the second duration threshold, and the second score threshold is smaller than the third score threshold.

In order to improve the sensitivity of model recognition in the above units, various different duration thresholds and score thresholds are set. This application sets three duration thresholds and three score thresholds. In the above device, by comparing the duration of the audio clip and the relationship between the highest recognition score and different corresponding thresholds, it is possible to first determine the candidate unknown character, that is, the possible unknown character, and then wait for the subsequent determination, which can improve the accuracy of the recognition model. Therefore, there are three situations in which it can be considered that the role corresponding to the audio is a candidate unknown role, that is, a possible unknown role. In the first case: the audio duration of at least one audio segment is less than the first duration threshold and the highest recognition score is less than the first score threshold; the second Two cases: the audio duration of at least one audio clip is greater than or equal to the first duration threshold and less than the third duration threshold and the highest recognition score is greater than or equal to the first score threshold and less than the second score threshold; the third case: at least one audio The audio duration of the segment is greater than or equal to the third duration threshold and less than the second duration threshold and the highest recognition score is greater than or equal to the second score threshold and less than the third score threshold.

In an embodiment of the present application, on the basis of including the first acquisition unit, the second acquisition unit, the determination unit and the registration unit, the specific determination unit is further refined, and the unit specifically includes: a fourth determination module , for determining that the character corresponding to the at least one audio segment is the unknown character when the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is smaller than the score threshold, that is, simultaneously satisfying When the audio duration of at least one audio clip is greater than or equal to the above-mentioned second duration threshold and the above-mentioned highest recognition score is less than the above-mentioned score threshold, it can be determined that the role corresponding to at least one audio clip is an unknown character; When the audio duration of at least one audio clip is greater than or equal to the second duration threshold and the highest recognition score is greater than or equal to the above score threshold, it is determined that the role corresponding to the at least one audio segment is a known role, that is, at least one audio segment is satisfied at the same time. When the audio duration of the segment is greater than or equal to the second duration threshold and the highest recognition score is greater than the score threshold, it may be determined that the character corresponding to at least one audio segment is a known character.

Through the above units, it can be determined whether the character corresponding to at least one audio segment is known or unknown, and the recognition of the unknown character is realized. After it is determined that the character corresponding to the audio is a known character or an unknown character, the automatic speech recognition technology can also be called to convert the speech into text or call the machine translation technology. In addition, it can also display the audio role name and the result of audio conversion to text, and then realize the functions of audio and processing result comparison and playback, audio clip selection, and audio role name modification.

In the actual voiceprint recognition process, the audio can be muted. The purpose is to identify the silent segment in the audio. Only when the audio is not silent can the subsequent role separation and role switching recognition be continued, so as to improve the efficiency of audio recognition. effect. In an embodiment of the present application, on the basis of including the first acquisition unit, the second acquisition unit, the determination unit and the registration unit, the above-mentioned first acquisition unit is refined, and the first acquisition unit includes the sixth determination unit module, a seventh determination module, and an eighth determination module, wherein the sixth determination module is configured to obtain the duration of the at least one audio segment when the first recognition result indicates that the at least one audio segment is the target mute segment, and When the duration of the at least one audio segment is greater than the fourth duration threshold, it is determined that the at least one audio segment is the target mute segment, and the corresponding role is empty, that is, at least one audio segment is considered to be mute and has no corresponding role; The seventh determination module is used to obtain the subsequent audio segment of the at least one audio segment when the duration of the at least one audio segment is less than or equal to the fourth duration threshold, obtain a second updated audio segment, and perform an update on the second updated audio segment. Perform the above-mentioned voiceprint recognition on the fragment to obtain a third recognition result, and if the above-mentioned third recognition result indicates that the duration of the above-mentioned updated audio is greater than the above-mentioned fourth duration threshold, determine that the above-mentioned at least one audio fragment is the above-mentioned target mute segment, that is, determine the mute After the segment, obtain the subsequent audio segment, identify the subsequent audio segment, and determine the mute segment or the non-mute segment according to the recognition result of the subsequent audio segment; the eighth determination module is used to represent the above-mentioned second update in the above-mentioned third recognition result If the audio duration of the audio segment is less than or equal to the fourth duration threshold, the fourth determination module is repeatedly executed until it is determined that the second updated audio segment is the target silent segment or the non-target silent segment. In the device, the target silent segment is determined first, and then the subsequent audio segment is obtained for determination, so that the result of determining whether the target silent segment is obtained is more accurate.

In one embodiment of the present application, on the basis of including the above-mentioned sixth determination module, seventh determination module and eighth determination module, the above-mentioned sixth determination module is refined, and this module includes: a fourth determination sub-module and a fifth determining submodule, wherein the fourth determining submodule is configured to determine that the at least one audio segment is the target mute segment when the duration of the at least one audio segment is greater than the third duration threshold, that is, the target mute is determined After the segment, if the duration is greater than the third duration threshold, it can be determined as the target mute segment; the fifth determining submodule is used to determine when the duration of the at least one audio segment is less than or equal to the third duration threshold and greater than the fourth duration In the case of a threshold, it is determined that the at least one audio segment is the target silence segment, and the third duration threshold is greater than the fourth duration threshold, that is, after the target silence segment is determined, the duration is less than or equal to the third duration threshold and greater than the above In the case of the fourth duration threshold, it may be determined as the target silence segment. Through the above device, the setting of the fourth duration threshold can eliminate the silence of the speaker in the audio, and can make the voiceprint recognition result more accurate.

In the actual use process, there are still cases where the speaker changes, so it is also necessary to perform role switching detection on the audio. In another embodiment of the present application, on the basis of including the first acquisition unit, the second acquisition unit, the determination unit and the registration unit, it also includes a ninth determination module, a tenth determination module, an eleventh determination module and a Twelve determination modules, wherein the ninth determination module is used to determine whether the historical role is the same as the current role when the historical role is not empty, wherein the current role is the role corresponding to the current at least one audio segment, and the above-mentioned The historical role is the role corresponding to the audio clip before the at least one audio clip, that is, first determine whether the role corresponding to the at least one audio clip is the same as the previous audio clip; In this case, it is determined that role switching has not occurred, that is, if at least one audio segment is the same as the role corresponding to the previous audio segment, then no role switching has occurred; In the same case, it is determined whether the duration of the at least one audio segment is greater than or equal to the third duration threshold, and if the duration of the at least one audio segment is greater than or equal to the third duration threshold, it is determined that the above-mentioned role switching occurs, that is, determining When the role corresponding to at least one audio clip is not the same as that of the previous audio clip, if the duration is greater than or equal to the above-mentioned third duration threshold, it is determined that a role switch has occurred; When the duration of the segment is less than the third duration threshold, acquire the subsequent audio segment of the at least one audio segment to obtain the third updated audio segment, and repeat the ninth determination module to the eleventh determination module at least once until Until the above-mentioned role switching occurs or does not occur, in the process of repeated execution, the above-mentioned current role is the role corresponding to the above-mentioned third updated audio segment, that is, after the role switching is determined, the subsequent audio segment is obtained, and the subsequent Identify the audio clips, and determine whether role switching occurs according to the recognition results of the subsequent audio clips. In the device, by first determining whether role switching occurs, and then determining whether the role switching occurs after obtaining subsequent audio clips, the result of determining whether role switching occurs is more accurate. Through this device, it is possible to accurately identify whether role switching occurs in the audio.

The above-mentioned audio processing device includes a processor and a memory, and the above-mentioned first acquisition unit, second acquisition unit, determination unit and registration unit are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding function.

The processor includes a kernel, and the kernel fetches corresponding program units from the memory. The kernel can be set to one or more, and the unknown character corresponding to the audio can be recognized by adjusting the kernel parameters.

Memory may include non-permanent memory in computer-readable media, random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), memory includes at least one memory chip.

An embodiment of the present invention provides a processor, where the processor is used to run a program, wherein the audio processing method is executed when the program is running.

An embodiment of the present invention provides a device. The device includes a processor, a memory, and a program stored on the memory and operable on the processor. When the processor executes the program, at least the following steps are implemented:

Step S101, acquiring at least one audio clip, and using a voiceprint recognition model to perform voiceprint recognition on the at least one audio clip to obtain a first recognition result;

Step S102, when the above-mentioned first recognition result indicates that the above-mentioned at least one audio segment is a non-target silent segment and the duration of the above-mentioned at least one audio segment is greater than or equal to a first duration threshold, obtain the highest recognition score in the above-mentioned first recognition result ;

Step S103, when the audio duration of the at least one audio segment is greater than or equal to the duration threshold and the highest recognition score is less than the score threshold, determine that the character corresponding to the at least one audio segment is an unknown character;

Step S104, registering the above-mentioned unknown character into the above-mentioned voiceprint recognition model library.

The devices in this article can be servers, PCs, PADs, mobile phones, etc.

The present application also provides a computer program product, which, when executed on a data processing device, is adapted to execute a program initialized with at least the following method steps:

Step S101, acquiring at least one audio clip, and using a voiceprint recognition model to perform voiceprint recognition on the at least one audio clip to obtain a first recognition result;

Step S102, when the above-mentioned first recognition result indicates that the above-mentioned at least one audio segment is a non-target silent segment and the duration of the above-mentioned at least one audio segment is greater than or equal to a first duration threshold, obtain the highest recognition score in the above-mentioned first recognition result ;

Step S103, when the audio duration of the at least one audio segment is greater than or equal to the duration threshold and the highest recognition score is less than the score threshold, determine that the character corresponding to the at least one audio segment is an unknown character;

Step S104, registering the above-mentioned unknown character into the above-mentioned voiceprint recognition model library.

In the above-mentioned embodiments of the present invention, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed technical content can be realized in other ways. Wherein, the device embodiments described above are only illustrative. For example, the division of the above-mentioned units can be a logical function division. In actual implementation, there may be another division method, for example, multiple units or components can be combined or integrated. to another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of units or modules may be in electrical or other forms.

The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the above integrated units are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the above-mentioned methods in various embodiments of the present invention. The aforementioned storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes. .

From the above description, it can be seen that the above-mentioned embodiments of the present application have achieved the following technical effects:

1), in the audio processing method of the present application, first obtain at least one audio segment, and use a voiceprint recognition model to perform voiceprint recognition on at least one audio segment to obtain a first recognition result, and then characterize at least one audio segment in the first recognition result When the audio segment is a non-target silent segment and the duration of at least one audio segment is greater than or equal to the first duration threshold, obtain the highest recognition score in the first recognition result; if the audio duration of at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold, it is determined that the role corresponding to at least one audio clip is an unknown character, and the second duration threshold is greater than the first duration threshold; finally, the unknown character is registered in the library of the voiceprint recognition model. By setting the recognition score threshold and the audio duration threshold, and judging the audio, the unknown characters appearing in the audio are detected, and the detected unknown characters are registered in the voiceprint recognition model library.

2) In the audio processing device of the present application, the first acquisition unit is used to acquire at least one audio clip, and use the voiceprint recognition model to perform voiceprint recognition on at least one audio clip to obtain a first recognition result, at least one audio clip A segment may be one audio segment or multiple audio segments, and in different application scenarios, the number of audio segments may be different. The second acquisition unit is used to obtain the highest recognition score in the first recognition result when the first recognition result indicates that at least one audio segment is a non-target silent segment and the duration of the at least one audio segment is greater than or equal to the first duration threshold; If at least one audio segment is a non-target silent segment, it means that the at least one audio segment is not a silent segment, that is, the highest recognition score is obtained only when at least one audio segment is not a silent segment. In addition, if the duration of at least one audio segment is too short and less than the first duration threshold, the determined character may be inaccurate. Therefore, another prerequisite for obtaining the highest recognition score is that the duration of at least one audio segment is greater than or equal to the first duration threshold. duration threshold. The determination unit is used to determine that the character corresponding to the at least one audio segment is an unknown character when the audio duration of at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold, and the second duration threshold is greater than the second duration threshold. A duration threshold; if the audio clip is too short, it cannot be accurately determined whether the corresponding character is an unknown character. The registration unit is used to register the above-mentioned unknown character into the above-mentioned voiceprint recognition model library. In the above audio processing device, by comparing the duration of the audio clip and the relationship between the highest recognition score and the corresponding threshold, it can be determined whether the character corresponding to the audio is an unknown character, and if it is determined to be an unknown character, register the unknown character in the voiceprint In the identified model library, voiceprint role separation can be performed later without pre-registration, thereby solving the problem in the prior art that requires pre-registration to perform voiceprint role separation. Compared with the prior art, this solution requires pre-registration For the solution, it is easy to use and applicable to a wide range of scenarios.

The foregoing are only preferred embodiments of the present application, and are not intended to limit the present application. For those skilled in the art, various modifications and changes may be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims (10)

1. An audio processing method, comprising:

acquiring at least one audio clip, and performing voiceprint recognition on the at least one audio clip by adopting a voiceprint recognition model to obtain a first recognition result;

under the condition that the first identification result represents that the at least one audio segment is a non-target mute segment and the duration of the at least one audio segment is greater than or equal to a first duration threshold, acquiring the highest identification score in the first identification result;

determining that the role corresponding to the at least one audio clip is an unknown role under the condition that the audio duration of the at least one audio clip is greater than or equal to a second duration threshold and the highest identification score is less than a score threshold, wherein the second duration threshold is greater than the first duration threshold;

registering the unknown character into the voiceprint recognized model library.

2. The method of claim 1, wherein determining that the corresponding role of the at least one audio piece is an unknown role in the case that the audio duration of the at least one audio piece is greater than or equal to a second duration threshold and the highest identification score is less than the score threshold comprises:

a first determination step, in which a role corresponding to the at least one audio clip is determined to be a candidate unknown role under the condition that the audio duration of the at least one audio clip is smaller than the second duration threshold and the highest identification score is smaller than the score threshold;

a second determining step, in which a subsequent audio clip of the at least one audio clip is obtained to obtain a first updated audio clip, the voiceprint recognition is performed on the first updated audio clip to obtain a second recognition result, the candidate unknown role is updated to be the known role under the condition that the second recognition result represents that the audio duration of the first updated audio clip is greater than or equal to the second duration threshold and the highest recognition score is greater than the score threshold, and the candidate unknown role is updated to be the unknown role under the condition that the second recognition result represents that the audio duration of the first updated audio clip is greater than or equal to the second duration threshold and the highest recognition score is less than or equal to the score threshold;

and under the condition that the second identification result represents that the audio duration of the first updated audio clip is smaller than the second duration threshold, repeatedly executing the second determining step until the role corresponding to the first updated audio clip is determined to be the known role or the unknown role.

3. The method of claim 2, wherein determining that the role corresponding to the at least one audio piece is a candidate unknown role in the case that the audio duration of the at least one audio piece is less than the second duration threshold and the highest recognition score is less than the score threshold comprises:

determining the role corresponding to the at least one audio clip as the candidate unknown role under the condition that the audio duration of the at least one audio clip is smaller than a first duration threshold and the highest identification score is smaller than a first score threshold;

determining that the role corresponding to the at least one audio clip is the unknown candidate role under the conditions that the audio duration of the at least one audio clip is greater than or equal to the first duration threshold and less than a third duration threshold and the highest identification score is greater than or equal to the first score threshold and less than a second score threshold, wherein the first duration threshold is less than the third duration threshold and the first score threshold is less than the second score threshold;

determining that the role corresponding to the at least one audio clip is the candidate unknown role under the condition that the audio duration of the at least one audio clip is greater than or equal to the third duration threshold and less than a second duration threshold and the highest identification score is greater than or equal to the second score threshold and less than a third score threshold, wherein the third duration threshold is less than the second duration threshold and the second score threshold is less than the third score threshold.

4. The method of claim 1, wherein determining that the corresponding role of the at least one audio segment is an unknown role in the case that the audio duration of the at least one audio segment is greater than or equal to the second duration threshold and the highest recognition score is less than the score threshold comprises:

determining that the role corresponding to the at least one audio clip is the unknown role under the condition that the audio duration of the at least one audio clip is greater than or equal to the second duration threshold and the highest identification score is less than the score threshold;

and under the condition that the audio duration of the at least one audio clip is greater than or equal to the second duration threshold and the highest identification score is greater than or equal to the score threshold, determining that the role corresponding to the at least one audio clip is a known role.

5. The method of claim 1, wherein obtaining at least one audio segment and performing voiceprint recognition on the at least one audio segment by using a voiceprint recognition model to obtain the first recognition result comprises:

a third determining step, in which, when the first recognition result indicates that the at least one audio clip is the target silent clip, the duration of the at least one audio clip is obtained, and when the duration of the at least one audio clip is greater than a fourth duration threshold, the at least one audio clip is determined to be the target silent clip, and the corresponding role is null;

a fourth determining step, in which, when the duration of the at least one audio clip is less than or equal to the fourth duration threshold, a subsequent audio clip of the at least one audio clip is obtained to obtain a second updated audio clip, the voiceprint recognition is performed on the second updated audio clip to obtain a third recognition result, and when the third recognition result indicates that the duration of the second updated audio clip is greater than the fourth duration threshold, the at least one audio clip is determined to be the target silent clip;

and in the case that the third identification result indicates that the audio duration of the second updated audio segment is less than or equal to the fourth duration threshold, repeating the fourth determining step until the second updated audio segment is determined to be the target mute segment or the non-target mute segment.

6. The method of claim 5, wherein determining that the at least one audio segment is the target silence segment if the duration of the at least one audio segment is greater than the fourth duration threshold comprises:

determining the at least one audio segment as the target silence segment when the duration of the at least one audio segment is greater than a second duration threshold;

and under the condition that the duration of the at least one audio segment is less than or equal to the second duration threshold and greater than the fourth duration threshold, determining that the at least one audio segment is the target silence segment, wherein the second duration threshold is greater than the fourth duration threshold.

7. The audio processing method of claim 1, further comprising:

a fifth determining step of determining whether the historical role is the same as the current role or not under the condition that the historical role is not empty, wherein the current role is a role corresponding to the current at least one audio clip, and the historical role is a role corresponding to an audio clip before the at least one audio clip;

a sixth determination step of determining that role switching has not occurred in a case where the historical role is the same as the current role;

a seventh determining step of determining whether the duration of the at least one audio clip is greater than or equal to a second duration threshold value or not when the historical role is different from the current role, and determining that the role switching occurs when the duration of the at least one audio clip is greater than or equal to the second duration threshold value;

and under the condition that the duration of the at least one audio clip is smaller than the second duration threshold, acquiring a subsequent audio clip of the at least one audio clip to obtain a third updated audio clip, and sequentially and repeatedly executing the fifth determining step to the seventh determining step at least once until the role switching is determined to occur or not to occur, wherein in the repeated executing process, the current role is a role corresponding to the third updated audio clip.

8. An audio processing apparatus, characterized in that the processing apparatus comprises:

the first acquisition unit is used for acquiring at least one audio clip and carrying out voiceprint recognition on the at least one audio clip by adopting a voiceprint recognition model to obtain a first recognition result;

the second obtaining unit is used for obtaining the highest identification score in the first identification result under the condition that the first identification result represents that the at least one audio segment is a non-target mute segment and the duration of the at least one audio segment is greater than or equal to a first duration threshold;

the determining unit is used for determining that the role corresponding to the at least one audio clip is an unknown role under the condition that the audio duration of the at least one audio clip is greater than or equal to a second duration threshold and the highest identification score is less than a score threshold, wherein the second duration threshold is greater than the first duration threshold;

and the registration unit is used for registering the unknown role in the voiceprint recognition model library.

9. A processor, characterized in that the processor is configured to run a program, wherein the program when running performs the method of any of claims 1 to 7.

10. An audio processing system, comprising: one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing the method of any of claims 1-7.

Images