Presentation 2010-12-21
Non-negative matrix factorization of segmental STRAIGHT speech spectrograms
Makoto KOSEKI, Kazunori MANO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we propose a new framework of non-negative matrix factorization (NMF) applied to segmental speech for spectrogram compression in STRAIGHT. Non-negative matrix factorization is a matrix decomposition method using non-negative constraints, and it is possible to extract various characteristic features as the form of decomposed matrix by controlling the initial values and the constraints of optimization measures. If an NMF is applied to a whole sentence, higher levels of some specific phoneme spectra would be involuntarily biased. In this case, although the higher-level spectra can be well approximated, the lower-level spectra may not be well approximated. The proposed method performs NMF based on the characteristics of speech segments. It is shown that the obtained spectrogram model provides better basis spectra for each cluster than normal NMF and mel-cepstral representations.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) STRAIGHT / Non-negative matrix factorization / speech segment
Paper # NLC2010-27,SP2010-100
Date of Issue

Conference Information
Committee NLC
Conference Date 2010/12/13(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Non-negative matrix factorization of segmental STRAIGHT speech spectrograms
Sub Title (in English)
Keyword(1) STRAIGHT
Keyword(2) Non-negative matrix factorization
Keyword(3) speech segment
1st Author's Name Makoto KOSEKI
1st Author's Affiliation College of Systems Engineering and Science, Shibaura Institute of Technology()
2nd Author's Name Kazunori MANO
2nd Author's Affiliation College of Systems Engineering and Science, Shibaura Institute of Technology
Date 2010-12-21
Paper # NLC2010-27,SP2010-100
Volume (vol) vol.110
Number (no) 356
Page pp.pp.-
#Pages 6
Date of Issue