Voice wake-up model training method and device, equipment and readable storage medium

Abstract

The invention discloses a voice wake-up model training method and device, equipment and a readable storage medium, and relates to the technical field of model training. Comprising the following steps: acquiring customized entries uploaded by a user, and determining a pronunciation sequence according to the customized entries; according to the customized vocabulary entry and the pronunciation sequence, searching and obtaining an audio corpus from a storage system; the audio corpus at least comprises a clean corpus, a synthetic corpus and a self-recording corpus; performing noise reverberation simulation processing on the audio corpus to obtain a simulation corpus carrying noise and reverberation; and training to obtain a voice wake-up model based on the simulation corpus. According to the scheme of the invention, the recognition of the specific voice instruction is realized, and the recognition accuracy in a noise environment is improved.

Claims

1. A method for training a voice wake-up model, characterized in that: The method comprises: Obtaining a custom word uploaded by a user, and determining a pronunciation sequence according to the custom word; According to the customized word entry and the pronunciation sequence, searching and acquiring audio corpus from a storage system; the audio corpus at least includes clean corpus, synthetic corpus and self-recorded corpus; Performing noise-reverberation simulation processing on the audio corpus to obtain a simulation corpus carrying noise and reverberation; Based on the simulation corpus, a speech wake-up model is trained.

2. The method for training a voice wake-up model according to claim 1, wherein: Performing noise and reverberation simulation processing on the audio corpus to obtain a simulation corpus carrying noise and reverberation, including: Get environmental reverberation, environmental noise, and specific noise generated by specific devices uploaded by users; The environmental reverberation, environmental noise and specific noise are added to the audio corpus to obtain a simulated corpus carrying noise and reverberation.

3. The method for training a voice wake-up model according to claim 2, wherein: Adding the environmental reverberation, environmental noise and specific noise to the audio corpus to obtain a simulated corpus carrying noise and reverberation, including: Obtain the user's noise requirements; According to the noise requirement, setting the ratio information of the environmental reverberation, environmental noise and specific noise; According to the ratio information, the environmental reverberation, environmental noise and specific noise are added to the audio corpus to obtain a simulated corpus carrying noise and reverberation.

4. The method for training a voice wake-up model according to claim 1, wherein: Obtaining a custom word uploaded by a user, and determining a pronunciation sequence according to the custom word, including: Obtaining a custom entry uploaded by a user, and determining candidate pronunciations of each character in the custom entry; Determine the polyphonetic characters in the customized entry, and delete the non-target pronunciations of the polyphonetic characters from the candidate pronunciations to obtain the target pronunciations of each character in the customized entry; The pronunciation sequence is generated based on the target pronunciation.

5. The method for training a voice wake-up model according to any one of claims 1 to 4, characterized in that: The method further comprises: Converting the voice wake-up model into a code library that can be integrated into a voice product software development kit; Integrate the code library into the voice product software development kit.

6. The method for training a voice wake-up model according to claim 5, characterized in that: The method further comprises: Burn the voice product software development kit into the relevant chip.

7. A training device for a voice wake-up model, characterized in that: The device comprises: A pronunciation determination module, used to obtain a custom word uploaded by a user and determine a pronunciation sequence according to the custom word; An audio search module, used to search and obtain audio corpus from a storage system according to the customized word entry and the pronunciation sequence; the audio corpus at least includes clean corpus, synthetic corpus and self-recorded corpus; A simulation processing module, used for performing noise and reverberation simulation processing on the audio corpus to obtain a simulation corpus carrying noise and reverberation; The model training module is used to train a speech wake-up model based on the simulation corpus.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the training method of the voice wake-up model described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the training method of the voice wake-up model described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the training method of the voice wake-up model described in any one of claims 1 to 6 is implemented.

Images