VOICE ACTIVITY DETECTION METHOD, RELATED DEVICE, AND APPARATUS

WIPO Patent Application WO/2019/101123

A1

A voice activity detection method comprises: receiving voice data comprising multiple frames of a voice signal (S200); determining an energy value and a spectral entropy value of one frame of the voice signal (S202); calculating a square root of the energy value of the voice signal, and/or calculating a square root of the spectral entropy value of the voice signal; determining a spectral entropy-energy square root of the voice signal according to at least one of the square root of the energy value and the square root of the spectral entropy value (S204); and if the spectral entropy-energy square root of the voice signal is less than a first predetermined threshold, determining that the voice signal is a non-voice frame, or if the spectral entropy-energy square root of the voice signal is not less than the first predetermined threshold, determining that the voice signal is a voice frame (S206). The voice activity detection method and a voice activity detection apparatus solve the technical problem of inaccurate voice segment detection in the prior art caused by solutions based on transient energy and spectral entropy in which threshold configuration is highly influenced by a recording environment, or by solutions based on spectral entropy and energy integrals in which a characteristic of a voice frame is not effectively reflected.

LIU JIZHONG (CN)

PCT/CN2018/116909

May 31, 2019

November 22, 2018

Click for automatic bibliography generation  

TENCENT TECH SHENZHEN CO LTD (CN)

G10L15/04

CN107731223A	2018-02-23
CN106920543A	2017-07-04
CN106356076A	2017-01-25
CN102436821A	2012-05-02
CN104464722A	2015-03-25
EP1569200A1	2005-08-31

DEQI INTELLECTUAL PROPERTY LAW CORPORATION (CN)

View/Download PDF      PDF Help

Previous Patent: METHOD AND DEVICE FOR TRANSMITTING DATA

Next Patent: MINIMIZATION OF DRIVE-TEST EXCEPTION CONFIGURATION PROCESSING METHOD AND DEVICE