Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
MODEL TRAINING METHOD AND APPARATUS, VOICE CONVERSION METHOD, DEVICE, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2022/121180
Kind Code:
A1
Abstract:
A voice conversion model training method and a voice conversion method, the training method comprising: acquiring sample audio and converting the sample audio into a sample mel spectrum (S101); collecting noise audio and jointly inputting the noise audio and the sample mel spectrum into a generative network to obtain an output mel spectrum (S102); inputting the output mel spectrum into a discriminative network to obtain the type probability of the output mel spectrum and a label of the output mel spectrum (S103); on the basis of the type probability of the output mel spectrum and the label of the output mel spectrum, implementing alternating iterative training of the generative network and the discriminative network, and using the trained generative network as a voice conversion model (S104), to thereby reduce the requirements of model building for audio corpora and reduce the complexity of model building.

Inventors:
CHEN MINCHUAN (CN)
MA JUN (CN)
WANG SHAOJUN (CN)
XIAO JING (CN)
Application Number:
PCT/CN2021/084219
Publication Date:
June 16, 2022
Filing Date:
March 31, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
PING AN TECH SHENZHEN CO LTD (CN)
International Classes:
G10L25/24; G10L21/013; G10L25/30
Domestic Patent References:
WO2019092931A12019-05-16
Foreign References:
CN112509600A2021-03-16
CN110136686A2019-08-16
CN110706692A2020-01-17
CN110136690A2019-08-16
CN109741736A2019-05-10
Attorney, Agent or Firm:
SHENZHEN ZHONGYI UNION INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: