Method for extracting short-time energy frequency value in voice endpoint detection

Abstract

The invention relates to voice detection technology in an automatic caption generating system, in particular to a method for extracting a short-time energy frequency value in voice endpoint detection. The method comprises the following steps: dividing an audio sampling sequence into frames with fixed lengths, and forming a frame sequence; extracting three audio characteristic parameters comprising short-time energy, short-time zero-crossing rate and short-time information entropy aiming at data of each frame; and calculating short-time energy frequency values of the data of each frame according to the audio characteristic parameters, and forming a short-time energy frequency value sequence. By combining the audio characteristic parameters of a time domain and a frequency domain, the method can develop respective advantages of the characteristic parameters, and can avoid respective disadvantages to a certain extent at the same time so as to effectively treat background noise of various different types.

Description

technical field [0001] The invention relates to a speech detection technology in an automatic subtitle generation system, in particular to a method for extracting short-time energy-frequency values ​​in speech endpoint detection. Background technique [0002] Speech endpoint detection technology is a new field of speech technology research, which is applied in automatic subtitle generation system. The current subtitle production method first needs to prepare a subtitle manuscript. This subtitle manuscript refers to a text file written in advance before making a TV program, which records the title of the program, what the host wants to say, and what the interviewee said. words and other content. When making TV programs, editors add audio and video materials to the storyboard of non-linear editing software, and then edit them according to the gist of the program. Editing operations generally include modifying the position of the material, adding some special effects, adding ...

AI Technical Summary

Problems solved by technology

This method can accurately distinguish speech from noises such as car engines and door closing sounds, but it is less effective in distinguishing speech from music

[0008] No matter which audio parameters are used, traditional speech endpoint detection methods have great shortcomings in specific noise environments

For example, energy-based methods do not perform well in low SNR environments; information-entropy-based algorithms fail in music backgrounds

Images