Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
SPEECH SEPARATION METHOD AND APPARATUS, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2023/020500
Kind Code:
A1
Abstract:
A speech separation method and apparatus, and a storage medium. The method comprises: inputting mixed audio data of a time domain and image data into a first neural network for feature fusion, and outputting K first feature maps (1022); inputting a spectrogram of a frequency domain into a second neural network for feature separation, and outputting K second feature maps (1024); obtaining K spectrogram masks on the basis of the K first feature maps and the K second feature maps (1025); and finally, obtaining K pieces of separated independent audio data on the basis of the K spectrogram masks and the spectrogram (1026). In this way, during speech separation, the first neural network is introduced for multi-perception feature extraction to enhance speech features to obtain the K first feature maps, the second neural network is introduced to perform K component separation on the spectrogram of the mixed speech data to obtain the K second feature maps, and spectrogram mask prediction is performed by using the first feature maps and the second feature maps, thereby improving prediction accuracy and realizing effective separation of the mixed audio data.

Inventors:
LU HUIJUN (CN)
CAI DUNBO (CN)
QIAN LING (CN)
HUANG ZHIGUO (CN)
Application Number:
PCT/CN2022/112831
Publication Date:
February 23, 2023
Filing Date:
August 16, 2022
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
CHINA MOBILE SUZHOU SOFTWARE TECH CO LTD (CN)
CHINA MOBILE COMMUNICATIONS GROUP CO LTD (CN)
International Classes:
G10L21/0272; G10L21/028
Foreign References:
CN113035227A2021-06-25
CN109326302A2019-02-12
CN112634875A2021-04-09
US20200202869A12020-06-25
JP2020140050A2020-09-03
CN109525787A2019-03-26
US20180122403A12018-05-03
US20200335121A12020-10-22
Attorney, Agent or Firm:
CHINA PAT INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: