Voice audio segmenting method and device

Abstract

The invention provides a voice audio segmenting method and a voice audio segmenting device. The voice audio segmenting method comprises the following steps: carrying out framing on voice audio, thus generating a plurality of data frames, and carrying out detecting; and when detecting that non-voice data frames satisfying the preset conditions exist in the plurality of data frames, determining nodes for segmenting the voice audio according to the non-voice data frames. With the adoption of the technical scheme provided by the invention, on one hand, the voice audio segmenting efficiency can be improved, so that the efficiency in constructing a voice annotation library is improved, on the other hand, aiming at large audio files, the segmenting can be rapidly and effectively realized, the manual segmenting of a user is not needed, thus the labor cost is greatly reduced, and the use experiment of the user is promoted.

Description

technical field [0001] The present invention relates to the technical field of voice recognition, in particular, to a voice audio segmentation method and a voice audio segmentation device. Background technique [0002] In related technologies, the purpose of speech recognition is to enable computers to understand human language. In order to achieve the above goals, it must first be trained with a large amount of data. Therefore, a piece of speech and its corresponding content become training marked data. After a large amount of data is repeatedly iterated through a certain mathematical model, the computer can convert human speech into text, and then perform semantic understanding. Therefore, for computers, for any recognition algorithm, labeling databases is the key. [0003] When long-duration speech is used for word-level annotation, the computer needs to cluster and align features by itself, but long-duration audio is not conducive to alignment during computer training, t...

AI Technical Summary

Problems solved by technology

[0004] (1) if figure 1 As shown, the efficiency of saving a file by recording a sentence is too low;

[0005] (2) if figure 2 As shown, a large audio file is generated and then segmented. The segmentation method is still based on manual segmentation and audio annotation. Although the annotation accuracy is high and the usability rate is high, the efficiency is low, especially it needs to be Manual entry of content text, saving audio, naming and other complicated operations

Images