Voice acquisition method and device, computer equipment and storage medium

Abstract

The invention relates to a voice acquisition method and device, computer equipment and a storage medium. The voice acquisition method comprises the steps that an audio stream is received and divided into a plurality of voice segments; the voiceprint characteristics of the voice segments are extracted; and the voiceprint characteristics of the voice segments are compared in a paired mode to obtaina fraction matrix corresponding to the audio stream, and the fraction matrix is used for expressing the corresponding fractions of the similarity of paired voice fragments; the voiceprint characteristics are clustered according to the fraction matrix, and the voice fragment numbers corresponding to multiple voice sources are obtained; and the corresponding voice segment is extracted according to the voice source with the largest voice fragment number. Through the voice acquisition method, the phenomenon that later manual operation is increased due to the voice quality problem can be avoided, and the voice acquisition efficiency is improved.

Description

technical field [0001] The present application relates to the technical field of voice, in particular to a voice collection method, device, computer equipment and storage medium. Background technique [0002] With the development of speech technology, the requirements for speech are getting higher and higher. As the basis and important link of speech machine learning model training, speech collection requires strict control of speech quality. During the voice collection process, the voice quality is easily affected by various factors. For example, speech is mixed with noise. However, most speech segments in the noise-mixed speech are acceptable. However, in the traditional speech collection method, even if most of the speech segments in the speech mixed with noise are qualified, the entire speech will be deemed unqualified. In order to remove a small part of unqualified speech, manual editing is required, or the entire speech is discarded for re-collection, which is time-...

AI Technical Summary

Problems solved by technology

However, most of the speech fragments in the noise-mixed speech are qualified

However, in the traditional speech collection method, even if most of the speech segments in the speech mixed with noise are qualified, it will be considered as unqualified for the entire speech

In order to remove a small part of unqualified speech, manual editing is required, or the entire speech is discarded for re-collection, which is time-consuming and laborious, and there is a technical problem of low speech collection efficiency

Images