Title:
NOISE-TOLERANT VOICE RECOGNITION DEVICE AND METHOD, AND COMPUTER PROGRAM
Document Type and Number:
WIPO Patent Application WO/2020/017226
Kind Code:
A1
Abstract:
This noise-tolerant voice recognition device 180, which has a high voice recognition accuracy even when only a single channel voice signal is available, includes: a voice emphasis unit 202 which receives, as an input, a voice signal 112 that is a signal in which a noise signal overlaps a voice signal that is a target signal, and outputs an emphasis voice signal 112 obtained by emphasizing the voice signal with a prescribed voice emphasis method; an extended feature extraction unit 200 which receives, as inputs, the emphasis voice signal 203 and the voice signal 112, and extracts respective feature amounts; and a voice recognition unit 204 which converts utterance content of the voice signal into text by using the feature amounts and an acoustic model 206 for the same.
Inventors:
FUJIMOTO MASAKIYO (JP)
KAWAI HISASHI (JP)
KAWAI HISASHI (JP)
Application Number:
PCT/JP2019/024279
Publication Date:
January 23, 2020
Filing Date:
June 19, 2019
Export Citation:
Assignee:
NAT INST INF & COMM TECH (JP)
International Classes:
G10L15/20; G10L15/06; G10L15/16
Domestic Patent References:
WO2017135148A1 | 2017-08-10 |
Foreign References:
JP2001092491A | 2001-04-06 | |||
JP2015102806A | 2015-06-04 | |||
US20170256254A1 | 2017-09-07 |
Attorney, Agent or Firm:
SHIMIZU, Satoshi (JP)
Download PDF: