Voice endpoint detection model training method and device, and voice endpoint detection model using method and device

Abstract

The invention discloses a voice endpoint detection model training method and device, and a voice endpoint detection model using method and device. The training method comprises the steps: inputting atraining audio into a generalized voice endpoint detection model; detecting a plurality of audio events existing in the training audio through the generalized voice endpoint detection model, wherein the plurality of audio events include a human speaking event, a mute event and at least one noise event; obtaining a voice and non-voice distinguishing result of the plurality of audio events output bythe generalized voice endpoint detection model; calculating a loss function based on the audio event label of the training audio and the output of the generalized voice endpoint detection model; andoptimizing the generalized voice endpoint detection model by controlling the loss function. According to the scheme of the embodiment of the invention, different types in the non-voice part can be distinguished, the classification accuracy can be improved, and the noise is not easy to misjudge as the voice part.

Description

technical field [0001] The invention belongs to the field of language models, in particular to a method and device for training and using a speech endpoint detection model. Background technique [0002] In related technologies, there are various types of speech endpoint detection models, including speech endpoint detection models that use thresholds such as short-term energy and zero-crossing rate as speech and non-speech discrimination standards, and speech endpoints that use discriminative models such as neural networks to train Detector. [0003] On the one hand, the threshold discrimination method mainly uses indicators such as short-term energy and zero-crossing rate as the threshold. After extracting the acoustic features from the audio, these indicators are calculated for each frame or each segment, and then the audio is distinguished by whether it reaches the threshold. Speech and non-speech parts. [0004] On the other hand, the model discrimination method mainly ...

AI Technical Summary

Problems solved by technology

[0006] The main defect of the related technology is that it cannot distinguish between speech and non-speech in a noisy environment.

Some noise-adaptive technical methods require a lot of labeled data for training in order to be robust to some specific noises, and these labeled data are often difficult to obtain, or the number is not so large

Images