VOICE SIGNAL PROCESSING MODEL TRAINING METHOD, ELECTRONIC DEVICE, AND STORAGE MEDIUM

Title:

VOICE SIGNAL PROCESSING MODEL TRAINING METHOD, ELECTRONIC DEVICE, AND STORAGE MEDIUM

Document Type and Number:

WIPO Patent Application WO/2019/100998

Kind Code:

A1

Abstract:

A voice signal processing model training method, an electronic device, and a storage medium. The method comprises: obtaining a sample voice and determining a task input feature of each voice signal processing task of the sample voice (S100); determining a target training loss function according to a training loss function of each voice signal processing task (S110); and inputting the task input feature of each voice signal processing task into a multi-task neural network to be trained, using minimization of the target training loss function as a training target, and updating parameters of a sharing layer and each task layer of the multi-task neural network to be trained until the multi-task neural network to be trained is converged, thereby obtaining a voice signal processing model (S120). By means of a multi-task neural network, the computing complexity is reduced and the training efficiency of a voice signal processing model is improved.

Inventors:

CHEN LIANWU (CN)
YU MENG (CN)
LUO MIN (CN)
SU DAN (CN)

Application Number:

PCT/CN2018/115704

Publication Date:

May 31, 2019

Filing Date:

November 15, 2018

Export Citation:

Click for automatic bibliography generation Help

Assignee:

TENCENT TECH SHENZHEN CO LTD (CN)

International Classes:

G10L17/04; G10L15/16

Domestic Patent References:

WO2017083399A2

2017-05-18

Foreign References:

CN104881678A	2015-09-02
CN107357838A	2017-11-17
CN106355248A	2017-01-25

Other References: