A sample generation method, device, server and storage medium

Abstract

The present invention provides a sample generation method, device, server and storage medium, which determines the target phoneme sequence of each target audio sample related to the target wake-up word that has been selected from the audio sample set; the judgment is based on the first phoneme in all target phoneme sequences Whether the comparison result obtained by the total quantity and the total quantity of the second phoneme meets the preset sample generation condition; if the comparison result does not meet the sample generation condition, select the target audio sample related to the target wake-up word from the audio sample set; The currently selected phoneme sequence of the target audio sample is processed to generate the target phoneme sequence; if the comparison result satisfies the sample generation condition, the target phoneme sequence of each selected target audio sample is determined as the sample of the target wake-up word. Based on the present invention, the purpose of sample generation can be realized on the basis of reducing the time-consuming and labor cost of sample generation.

Description

A sample generation method, device, server and storage medium technical field The present invention relates to the technical field of voice wake-up, more specifically, to a kind of sample generation method, device, service device and storage medium. Background technique [0002] Voice wake-up is academically called keyword spotting (KWS for short), which refers to the fact that the user speaks the wake-up word by To wake up the electronic device, make the electronic device enter the state of waiting for the voice command or make the electronic device directly execute the predetermined voice instruction. The core content of the voice wake-up technology is the voice recognition model. Only by first using the voice recognition model to identify the The wake word can wake up the electronic device based on the recognized wake word. [0003] At present, the speech recognition model needs to use a large number of samples to train and generate, and the positive samples are ma...

AI Technical Summary

Problems solved by technology

[0004] This sample generation method needs to manually record a large number of positive samples related to the wake-up word whenever a new wake-up word appears, which has high labor costs and a long recording period; and, because the negative samples may include both the voice of the corresponding wake-up word Frames include speech frames that do not correspond to wake-up words, and the number of speech frames that do not correspond to wake-up words in negative samples is also uncertain, so it is necessary to repeatedly select negative samples so that the number of speech frames corresponding to wake-up words in all positive and negative samples and The number of speech frames that do not correspond to wake-up words meets a certain ratio, and the negative sample selection process takes a long time

Images