Phonetic end point detection method and device therefor

Abstract

The invention provides a phonetic end point detection method and a device therefor. The phonetic end point detection method comprises: receiving phonetic data with noise, dividing the phonetic data with noise into plural overlapped phonetic frames, carrying out rapid Fourier variation calculation to obtain the frequency spectrum of each phonetic frame; dividing the frequency spectrum of each phonetic frame into plural sub-bands uniformly, and generating sub-band power spectral entropy density according to sub-band energy; weighing the sub-band power spectral entropy density to obtain sub-band weight power spectral entropy of each phonetic frame; judging the current phonetic frame as a noise section or phonetic section according to preset phonetic end point judgment threshold and sub-band weight power spectral entropy. The invention adopts sub-band power spectral entropy as phonetic eigenvalue of VAD judgment, and adaptively selects sub-band number and weight factor according to actual application environment, thus improving accuracy and preciseness of phonetic detection and obviously enhancing signal-to-noise ratio of communication.

Description

technical field [0001] The invention relates to the technical field of voice recognition, in particular to a voice endpoint detection method and device. Background technique [0002] Realizing clear voice communication in a high-noise environment is an urgent problem to be solved by many scientists and engineers. Communication in a high-noise environment, the background noise interferes a lot with the voice signal, which can lead to unclear calls and low intelligibility in the communication system. Therefore, how to maintain a high-quality, high-definition communication system in a complex background noise environment is particularly important. [0003] In the speech system, background noise is often input together with the speech signal, so how to accurately judge the presence or absence of the speech signal in the input signal and determine its starting and ending positions becomes a problem of suppressing and removing speech noise. The key point of the voice endpoint de...

AI Technical Summary

Problems solved by technology

[0006] The embodiment of the present invention provides a speech endpoint detection method and device, which are used to solve the problem in the prior art that speech recognition is easily polluted by noise and cause performance degradation in a low signal-to-noise ratio and complex noise environment, and to achieve high-decibel noise Accurate recognition of speech in the environment, the function of suppressing the interference of background noise on the communication system, and improving the accuracy of speech detection technology in various noise environments

Images