首页>
外国专利>
COMPUTER SYSTEM CREATING SPEAKER ADAPTATION WITHOUT TEACHER IN DNN-BASED SPEECH SYNTHESIS, AND METHOD AND PROGRAM EXECUTED IN COMPUTER SYSTEM
COMPUTER SYSTEM CREATING SPEAKER ADAPTATION WITHOUT TEACHER IN DNN-BASED SPEECH SYNTHESIS, AND METHOD AND PROGRAM EXECUTED IN COMPUTER SYSTEM
A computer system 1 includes a speaker information estimation unit 130 that estimates the speaker information of an unknown speaker on the basis of the acoustic feature amount for the unknown speaker without the need to enter text as teacher data. The speaker information of unknown speaker includes a speaker code that represents similarity by probability between a distribution of the acoustic feature amount for the unknown speaker and a distribution for each of the acoustic feature amounts for a plurality of known speakers. The computer system 1 further comprises: a synthesized acoustic feature amount generation unit 220 for generating a synthesized acoustic feature amount for the unknown speaker on the basis of a language feature amount for an input text and the speaker information of the unknown speaker, using acoustic models (DNN) 230 of multiple speakers; and a synthesized speech generation unit 240 for generating a synthesized speech of the unknown speaker on the basis of the synthesized acoustic feature amount of the unknown speaker.
展开▼
