Title:
SPEECH PROCESSING DEVICE, SPEECH PROCESSING METHOD, AND RECORDING MEDIUM
Document Type and Number:
WIPO Patent Application WO/2018/051945
Kind Code:
A1
Abstract:
It is difficult to analyze (interpret) an impact that each element of a feature vector extracted from a speech signal received has on the results of speaker recognition for the speech signal. A speech processing device according to the present invention includes: an acoustic model storage unit that stores one or more acoustic models; an acoustic statistics calculation unit that calculates acoustic features from the received speech signal, and calculates acoustic diversity represented by a vector exhibiting degree of variation in sound types using the calculated acoustic features and the stored acoustic models; a partial feature extraction unit that calculates weighted acoustic diversity using the calculated acoustic diversity and a selection coefficient, and calculates a recognition feature amount serving as the information for recognizing information indicating the identity and language of a speaker associated with the speech signal using the calculated weighted acoustic diversity and the acoustic features; and a partial feature integration unit that calculates a feature vector using the calculated recognition feature amount.
Inventors:
YAMAMOTO HITOSHI (JP)
KOSHINAKA TAKAFUMI (JP)
SUZUKI TAKAYUKI (JP)
KOSHINAKA TAKAFUMI (JP)
SUZUKI TAKAYUKI (JP)
Application Number:
PCT/JP2017/032666
Publication Date:
March 22, 2018
Filing Date:
September 11, 2017
Export Citation:
Assignee:
NEC CORP (JP)
International Classes:
G10L17/00; G10L15/02; G10L15/10
Foreign References:
JP2016075740A | 2016-05-12 | |||
JP2016061824A | 2016-04-25 |
Attorney, Agent or Firm:
SHIMOSAKA Naoki (JP)
Download PDF:
Previous Patent: PEOPLE FLOW ESTIMATION DEVICE, PEOPLE FLOW ESTIMATION METHOD, ABD RECORDING MEDIUM
Next Patent: INTEGRATED CIRCUIT
Next Patent: INTEGRATED CIRCUIT