Audio extraction method, device, equipment and readable storage medium

Abstract

The present invention provides an audio extraction method, device, equipment and readable storage medium. In the solution provided by the present invention, in processing the audio to be processed, the voice audio of the audio to be processed is first obtained, and then the obtained voice audio is segmented, and then it is identified whether the speech audio after segmentation is the target speech audio that matches the registered audio. Finally, the target speech audio obtained is the speech audio after segmentation. The speech audio is speech audio with segmentation and can identify sentence pauses, so that the output speech audio is more in line with the user's speaking rules.

Description

Technical Field

[0001] The present invention relates to the field of audio processing technology, and in particular to an audio extraction method, apparatus, device and readable storage medium. Background Art

[0002] Audio processing is a function available on many existing devices. It primarily isolates the target speaker's voice from a stream of collected speech signals. This function is required in a variety of speech signal processing scenarios and has garnered increasing attention in recent years. For example, extracting the target speaker's voice can be performed in everyday speech scenarios, such as a phone conversation, a meeting, a broadcast, or a television program.

[0003] In existing solutions, extracting the target subject's speech primarily relies on a voiceprint recognition model. First, a segment of the target speaker's audio is required as registration audio. The measured speech stream is first processed by a VAD (Voice Activity Detection) module to detect which frames contain human voice and which frames are silent (silence or background sound), thereby dividing the measured speech stream into segments. These segments are then segmented according to a specific window length and frame shift, typically around 1 second and 0.5 seconds. The voiceprint recognition model then compares the voiceprint with the registration audio, generating a score for each frame. The voiceprint model then extracts voiceprint features from the registration audio and each frame through forward computation. The model then calculates the cosine score or PLDA (Probabilistic Linear Discriminant) score for these features. A higher score indicates a greater likelihood that the two speech segments belong to the same speaker. Finally, the speech frames that exceed the threshold are concatenated to form the separation result.

[0004] Existing solutions are limited by VAD performance and the minimum discrimination time for voiceprint recognition. They often discard a significant amount of the target speaker's audio at speaker change points, resulting in abrupt interruptions at the end of the segmented sentences, which sounds unnatural. For example, when the speaker speaks quickly or there are short pauses between alternating sentences, VAD often fails to detect pauses. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide an audio extraction method, apparatus, device, and readable storage medium to achieve audio extraction of specific person audio that can identify sentence pauses.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0007] An audio extraction method, comprising:

[0008] Get the voice audio in the audio to be processed;

[0009] Perform sentence end judgment on the speech audio to obtain at least one complete segment;

[0010] Perform voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio.

[0011] Optionally, in the above audio extraction method, obtaining the speech audio in the audio to be processed includes:

[0012] Get the audio to be processed;

[0013] Classifying the audio type in the audio to be processed;

[0014] Recognize the speech audio in the classification result.

[0015] Optionally, in the above audio extraction method, after performing sentence end determination on the speech audio to obtain at least one complete segment, the method further includes:

[0016] Get the time node corresponding to the recognized end of the sentence;

[0017] Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

[0018] Optionally, in the above audio extraction method, performing voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio includes:

[0019] Get the voiceprint features of the registered audio;

[0020] Extracting voiceprint features of the complete speech segment;

[0021] Determine whether the similarity between the voiceprint feature of the complete speech segment and the voiceprint feature of the registration audio is greater than a preset value, and if so, use the complete speech segment as the target speech audio;

[0022] Otherwise, the complete segment is discarded.

[0023] Optionally, the above audio extraction method further includes:

[0024] Based on the sequence of the time nodes of the recognized target voice audio, each target voice audio is spliced ​​together.

[0025] An audio extraction device, comprising:

[0026] A speech audio extraction unit, used to obtain speech audio from the audio to be processed;

[0027] a punctuation prediction model, configured to determine the end of a sentence on the speech audio to obtain at least one complete segment;

[0028] The voiceprint model is used to perform voiceprint recognition on the complete speech segment to obtain target speech audio that matches the registered audio.

[0029] Optionally, in the above-mentioned audio extraction device, the speech audio extraction unit includes:

[0030] A VAD module is used to obtain audio to be processed and classify the audio type in the audio to be processed;

[0031] The speech recognition module is used to recognize the speech audio in the classification results.

[0032] Optionally, in the audio extraction device, the punctuation prediction model, after performing sentence-end determination on the speech audio and obtaining at least one complete segment, is further used to:

[0033] Get the time node corresponding to the recognized end of the sentence;

[0034] The punctuation prediction model is also used to:

[0035] Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

[0036] An audio extraction device, comprising: a memory and a processor;

[0037] The memory is used to store programs;

[0038] The processor is used to execute the program to implement each step of the audio extraction method described above.

[0039] A readable storage medium stores a computer program thereon, wherein when the computer program is executed by a processor, the steps of any one of the above-mentioned audio extraction methods are implemented.

[0040] Based on the above technical solution, in the above solution provided by the embodiment of the present invention, in processing the audio to be processed, the voice audio of the audio to be processed is first obtained, and then the obtained voice audio is segmented, and then it is identified whether the speech audio after segmentation is the target speech audio that matches the registered audio. Finally, the target speech audio obtained is the speech audio after segmentation. The speech audio is a speech audio with segmentation, which can identify sentence pauses, so that the output speech audio is more in line with the user's speaking rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0042] Figure 1 A schematic diagram of the flow of the audio extraction method disclosed in the embodiment of the present application;

[0043] Figure 2 A schematic flow chart of an audio extraction method disclosed in another embodiment of the present application;

[0044] Figure 3 A schematic flow chart of an audio extraction method disclosed in another embodiment of the present application;

[0045] Figure 4 A schematic diagram of the structure of the audio extraction device disclosed in an embodiment of the present application;

[0046] Figure 5 A schematic structural diagram of an audio extraction device disclosed in another embodiment of the present application;

[0047] Figure 6 A schematic structural diagram of an audio extraction device disclosed in another embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] In order to solve the problem of audio information loss caused by the inability to detect sentence pauses when performing speech extraction in existing solutions, the present application discloses an audio extraction method. Through this extraction method, the end of the sentence of the speech audio in the audio to be processed is judged, so that the recognized audio has a pause rhythm, thereby improving the reliability of the recognition result.

[0050] Figure 1 The audio extraction method disclosed in the embodiment of this application is an audio extraction method disclosed in this application. Figure 1 , the method comprising:

[0051] Step S101: Acquire speech audio from the audio to be processed.

[0052] In this solution, the audio to be processed may include not only speech audio but also various other interfering audio, such as the operating sound of a device, the hum of a car, background music, etc. In this solution, a speech recognition module can be used to identify the speech audio in this audio information. In this solution, the speech audio refers to the sound of a person speaking.

[0053] Step S102: performing sentence end determination on the speech audio to obtain at least one complete segment.

[0054] In this solution, after obtaining the voice audio, a punctuation prediction model is used to determine the sentence punctuation position in the voice audio, and then accurately determine the end of the sentence of the voice audio, and add corresponding punctuation marks at the sentence punctuation position to obtain a complete paragraph. A voice audio may contain multiple complete paragraphs. In this solution, the end of the sentence refers to the position that should be marked as a period, question mark or exclamation mark to represent the end of a sentence.

[0055] In this solution, the punctuation prediction model is a pre-trained artificial intelligence model. When performing punctuation prediction, the artificial intelligence model first converts the speech audio into recognizable characters, such as Chinese characters or English characters. The artificial intelligence model uses these recognized characters as input data, predicts punctuation marks, obtains the punctuation positions in these characters, uses the positions corresponding to the punctuation positions in the audio data as the sentence break positions, and adds corresponding punctuation marks at these positions to obtain a complete paragraph.

[0056] Step S103: performing voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio.

[0057] Since the audio to be processed may contain voice audio of different people at different time points, in this solution, through voiceprint recognition technology, it can be determined whether the obtained complete segment is the target voice audio that matches the registered audio. During the recognition process, if the voiceprint similarity between the complete segment and the registered audio is greater than a preset value, it indicates that the complete segment is the speech content of the target object and is recorded as the target voice audio.

[0058] In the technical solution disclosed in the embodiment of the present application, when processing the audio to be processed, the voice audio of the audio to be processed is first obtained, and then the obtained voice audio is segmented, and then it is identified whether the speech audio after segmentation is the target speech audio that matches the registered audio. Finally, the target speech audio obtained is the speech audio after segmentation, which is a speech audio with segmentation, so that the output speech audio is more in line with the user's speaking rules.

[0059] In another embodiment of the present application, the audio to be processed may include not only speech audio but also various other interference audios, such as the running sound of the device, the beeping sound of the car, background music, and other sounds emitted by non-target objects. In this solution, in order to better recognize speech audio, see Figure 2 , obtain the voice audio in the audio to be processed, which may include:

[0060] Step S201: Acquire audio to be processed.

[0061] Step S202: classify the audio types in the audio to be processed.

[0062] After acquiring the audio to be processed, the VAD module (Voice Activity Detection) classifies it. The VAD module first processes the audio stream to detect which frames contain human voices and which frames are silent or background. This classifies the audio stream into segments. These segments are then segmented using a specific window length and frame shift, typically around 1 second and 0.5 seconds.

[0063] Step S203: Identify the speech audio in the classified audio to be processed.

[0064] In this step, the classification results processed by the VAD module are recognized by the speech recognition module to obtain the speech audio in the classification results. The speech audio refers to the audio of the user speaking in the audio to be processed.

[0065] Therefore, the above solution can accurately eliminate the interfering audio in the audio to be processed, thereby improving the accuracy of audio processing.

[0066] In the technical solution disclosed in the embodiment of the present application, after performing voiceprint recognition on the speech audio of a complete sentence segment, the voiceprint recognition of the next segment can be continued, that is, performing voiceprint recognition on the complete sentence segment to obtain the target speech audio that matches the registered audio includes: obtaining the time node corresponding to the recognized end of the sentence; taking the time node as the starting time node, performing sentence end judgment on the speech audio to obtain at least one complete sentence segment.

[0067] Specifically, the time node corresponding to the end of the identified complete segment is obtained; taking the time node as the starting time node, the speech audio located after the time node in the speech audio is obtained for sentence end judgment, and the next complete segment located after the time node is obtained, until all sentence ends of the speech audio are judged.

[0068] In this solution, the sentence end judgment of the voice audio and the voiceprint recognition of the complete segment can be performed simultaneously. In this case, after obtaining the first complete segment, voiceprint recognition is performed on the complete segment. While voiceprint recognition is being performed, the sentence end judgment is continued for the voice audio after the time node corresponding to the sentence end of the voice audio. Alternatively, the sentence end judgment of the voice audio can be performed first. After the sentence end judgment is performed on all the voice audio, N complete segments are obtained. Then, voiceprint recognition is performed on the N complete segments in sequence according to the time nodes, and the target voice audio obtained by recognition is spliced ​​to obtain the complete target voice audio contained in the audio to be processed.

[0069] See also Figure 3 In the technical aspects disclosed in the embodiments of the present application, voiceprint recognition is performed on the complete speech segment to obtain a target voice audio that matches the registered audio, specifically including:

[0070] Step S301: Acquire the voiceprint features of the registered audio.

[0071] The registration audio is a speech audio of the target object, and the voiceprint features of the registration audio can be obtained by processing the registration audio.

[0072] Step S302: extracting the voiceprint features of the complete speech segment.

[0073] In this step, the voiceprint features of each complete speech segment are extracted. Different users have different voiceprint features. The voiceprint features can be compared to determine whether two voice audios are from the same user.

[0074] Step S303: Determine whether the similarity between the voiceprint feature of the complete speech segment and the voiceprint feature of the registered audio is greater than a preset value. If so, execute step S304.

[0075] In this step, the voiceprint features of the extracted complete speech segment are compared with the voiceprint features of the registered audio to determine whether the similarity is greater than a preset value. If the similarity between the two voiceprint features is greater than the preset value, it indicates that the two voices are from the same user; otherwise, they are from different users. The preset value can be set based on user needs.

[0076] Step S304: taking the complete segment as the target speech audio, extracting the next complete segment.

[0077] In this step, after the voiceprint features of a complete speech segment are compared with the voiceprint features of the registered audio, the next complete speech segment is extracted and the voiceprint features of the next complete speech segment are compared with the voiceprint features of the registered audio.

[0078] In this step, when the similarity between the voiceprint features of the compared complete segment and the voiceprint features of the registered audio is greater than a preset value, it indicates that the two are voice audios emitted by the same user. The voice audio of the complete segment is used as the target voice audio, and the next complete segment is extracted to compare the voiceprint features of the next complete segment with the voiceprint features of the registered audio.

[0079] Step S305: Otherwise, discard the complete segment and extract the next complete segment.

[0080] In this step, when the similarity between the voiceprint features of the compared complete segment and the voiceprint features of the registered audio is less than a preset value, it indicates that the two are voice audios emitted by different users. The voice audio of the complete segment is discarded, and the next complete segment is extracted to compare the voiceprint features of the next complete segment with the voiceprint features of the registered audio.

[0081] After the voiceprint features of all complete speech segments are compared with the voiceprint features of the registered audio, the identified target speech audio needs to be spliced ​​to obtain the coherent target speech audio contained in the audio to be processed. During the splicing process, each target speech audio can be spliced ​​based on the sequence of the time nodes of the identified target speech audio. The time node of the target speech audio can refer to the time node of the target speech audio in the audio to be processed, or it can refer to the time node when the complete speech segment is determined to be the target speech audio. Of course, each target speech audio can also be automatically spliced ​​in sequence based on the sequence of the determined target speech audio.

[0082] This embodiment discloses an audio extraction device. For the specific working contents of each unit in the device, please refer to the contents of the above method embodiment.

[0083] The audio extraction device provided by an embodiment of the present invention is described below. The audio extraction device described below and the audio extraction method described above can refer to each other.

[0084] See also Figure 4 The audio extraction device disclosed in the embodiment of the present application may include: a speech audio extraction unit A, a punctuation prediction model B and a voiceprint model C.

[0085] The speech audio extraction unit A corresponds to step S101 in the above method and is used to obtain the speech audio in the audio to be processed;

[0086] Punctuation prediction model B, corresponding to step S102 in the above method, is used to determine the end of a sentence on the speech audio to obtain at least one complete segment;

[0087] The voiceprint model C corresponds to step S103 in the above method and is used to perform voiceprint recognition on the complete speech segment to obtain a target speech audio that matches the registered audio.

[0088] See also Figure 5 , corresponding to the above method, the speech audio extraction unit A may include a VAD module A1 and a speech recognition module A2:

[0089] VAD module A1, used to obtain audio to be processed and classify the audio type in the audio to be processed;

[0090] The speech recognition module A2 is used to recognize the speech audio in the classification results.

[0091] Corresponding to the above method, after performing sentence-end judgment on the speech audio to obtain at least one complete segment, the punctuation prediction model B is further used to:

[0092] Get the time node corresponding to the recognized end of the sentence;

[0093] The punctuation prediction model is also used to:

[0094] Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

[0095] Figure 6 For a hardware structure diagram of the audio extraction device provided by an embodiment of the present invention, see Figure 6 As shown, it may include: at least one processor 100, at least one communication interface 200, at least one memory 300 and at least one communication bus 400;

[0096] In the embodiment of the present invention, the number of the processor 100, the communication interface 200, the memory 300, and the communication bus 400 is at least one, and the processor 100, the communication interface 200, and the memory 300 communicate with each other through the communication bus 400; obviously, Figure 6 The communication connections shown for the processor 100, communication interface 200, memory 300, and communication bus 400 are merely optional;

[0097] Optionally, the communication interface 200 may be an interface of a communication module, such as an interface of a GSM module;

[0098] The processor 100 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0099] The memory 300 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0100] The processor 100 is specifically configured to:

[0101] Get the voice audio in the audio to be processed;

[0102] Perform sentence end judgment on the speech audio to obtain at least one complete segment;

[0103] Perform voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio.

[0104] Corresponding to the above method, when the processor executes the acquisition of the voice audio in the audio to be processed, it is specifically used to execute:

[0105] Get the audio to be processed;

[0106] Classifying the audio type in the audio to be processed;

[0107] Recognize the speech audio in the classification result.

[0108] Corresponding to the above method, after performing sentence end determination on the voice audio to obtain at least one complete segment, the processor is further configured to execute:

[0109] Get the time node corresponding to the recognized end of the sentence;

[0110] Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

[0111] Corresponding to the above method, when the processor performs voiceprint recognition on the complete speech segment to obtain the target voice audio that matches the registration audio, it is specifically configured to perform:

[0112] Get the voiceprint features of the registered audio;

[0113] Extracting voiceprint features of the complete speech segment;

[0114] Determine whether the similarity between the voiceprint feature of the complete speech segment and the voiceprint feature of the registration audio is greater than a preset value, and if so, use the complete speech segment as the target speech audio;

[0115] Otherwise, the complete segment is discarded.

[0116] Corresponding to the above method, the processor is further configured to execute: splicing the target voice audios based on the order of the time nodes of the identified target voice audios.

[0117] The audio extraction device can be a voice recorder, a mobile phone or other smart terminals with audio processing functions.

[0118] For the above method, the present application also discloses a readable storage medium, which can be a chip with data storage function, and the chip stores preset computer programs. These computer programs can be triggered to execute. When the computer program is executed by the processor, the various steps of the audio extraction method described in any one of the above items are implemented.

[0119] For example, when a computer program is triggered, it is used to:

[0120] Get the voice audio in the audio to be processed;

[0121] Perform sentence end judgment on the speech audio to obtain at least one complete segment;

[0122] Perform voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio.

[0123] Corresponding to the above method, when the processor executes the acquisition of the voice audio in the audio to be processed, it is specifically used to execute:

[0124] Get the audio to be processed;

[0125] Classifying the audio type in the audio to be processed;

[0126] Recognize the speech audio in the classification result.

[0127] For the convenience of description, the above system is described as being divided into various modules according to their functions. Of course, when implementing the present invention, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0128] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0129] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0130] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0131] It should also be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0132] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An audio extraction method, characterized in that: include: Get the voice audio in the audio to be processed; Perform sentence end judgment on the speech audio to obtain at least one complete segment; The step of determining the end of a sentence on the speech audio to obtain at least one complete paragraph comprises: converting the speech audio into recognizable characters, inputting the characters into a punctuation prediction model, predicting punctuation marks, and obtaining punctuation positions in the characters; obtaining sentence break positions in the speech audio based on positions in the speech audio corresponding to the punctuation positions, and adding corresponding punctuation marks at the sentence break positions, wherein the sentence end refers to a position marked as a period, question mark, or exclamation mark to indicate the end of a sentence, to obtain the at least one complete paragraph; Performing voiceprint recognition on the complete speech segment to obtain target voice audio that matches the registered audio; Among them, the voiceprint recognition of the complete paragraph is performed to obtain the target voice audio that matches the registration audio, including: obtaining the voiceprint features of the registration audio, where the registration audio is a voice audio segment of the target object; extracting the voiceprint features of the complete paragraph; judging whether the similarity between the voiceprint features of the complete paragraph and the voiceprint features of the registration audio is greater than a preset value, and if so, taking the complete paragraph as the target voice audio; otherwise, discarding the complete paragraph.

2. The audio extraction method according to claim 1, wherein Get the voice audio in the audio to be processed, including: Get the audio to be processed; Classifying the audio type in the audio to be processed; Recognize the speech audio in the classification result.

3. The audio extraction method according to claim 2, wherein: Perform sentence end determination on the speech audio to obtain at least one complete segment, including: Get the time node corresponding to the recognized end of the sentence; Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

4. The audio extraction method according to claim 3, wherein: Also includes: Based on the order of the time nodes of the recognized target voice audio, each target voice audio is spliced.

5. An audio extraction device, characterized in that: include: A speech audio extraction unit, used to obtain speech audio from the audio to be processed; a punctuation prediction model, configured to determine the end of a sentence on the speech audio to obtain at least one complete segment; The punctuation prediction model is specifically used to: convert the speech audio into recognizable characters, input the characters into the punctuation prediction model, predict punctuation marks, and obtain punctuation positions in the characters; obtain sentence break positions in the speech audio based on positions in the speech audio corresponding to the punctuation positions, and add corresponding punctuation marks at the sentence break positions, where the end of a sentence refers to a position marked as a period, question mark, or exclamation mark to indicate the end of a sentence, to obtain the at least one complete paragraph; A voiceprint model is used to perform voiceprint recognition on the complete speech segment to obtain a target voice audio that matches the registered audio; Among them, the voiceprint model is specifically used to: obtain the voiceprint features of the registration audio, where the registration audio is a speech audio segment of the target object; extract the voiceprint features of the complete paragraph; determine whether the similarity between the voiceprint features of the complete paragraph and the voiceprint features of the registration audio is greater than a preset value. If it is greater than the preset value, the complete paragraph is used as the target speech audio; otherwise, the complete paragraph is discarded.

6. The audio extraction device according to claim 5, characterized in that The speech audio extraction unit comprises: A VAD module is used to obtain audio to be processed and classify the audio type in the audio to be processed; The speech recognition module is used to recognize the speech audio in the classification results.

7. The audio extraction device according to claim 6, characterized in that Punctuation prediction models are also used for: Get the time node corresponding to the recognized end of the sentence; The punctuation prediction model is also used to: Taking the time node as the starting time node, the voice audio is subjected to sentence end judgment to obtain at least one complete segment.

8. An audio extraction device, characterized in that include: including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the various steps of the audio extraction method according to any one of claims 1 to 4.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the audio extraction method according to any one of claims 1 to 4 is implemented.

Images