Presentation | 2006-12-15 Applying EM algorithm to 2ch BSS based on sparseness of speech Yosuke IZUMI, Hirokazu KAMEOKA, Nobutaka ONO, Shigeki SAGAYAMA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, we propose a new approach to sparseness-based BSS by applying EM algorithm. Based on sparseness of speech, under an observation model that only one of sources is active on each time-frequency component, the directions of ariival and the contribution ratio of each source at time-frequnecy components are estimated by EM algorithm. Our method has the following advantaages: 1) a common objective function is maximized in localization step and separation step, which are corresponding to E-step and M-step in EM algorithm, respectively, 2) it gives us a frame work to estimate the number of sources by information criterion since the objenction function is likelihood. 3) it enables to introduce physical observation model like the diffused sound field because the likelihood is defined as the original singal domain (time-frequency domain), not feature domain like time differecne or intensity ratio. 4) no heuristic of parameters are required since the magnutude of noise (variance) included in observation model can be also estimated by observed signals. Though our framework can be generally applied to N channel BSS, we derive an explicit algorithm with focusing 2ch BSS in this paper, and show some results by numerical simulation. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Blind Source Separation / EM algorithm / speech sparseness / time-frequency masking |
Paper # | EA2006-96 |
Date of Issue |
Conference Information | |
Committee | EA |
---|---|
Conference Date | 2006/12/8(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 | Engineering Acoustics (EA) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Applying EM algorithm to 2ch BSS based on sparseness of speech |
Sub Title (in English) | |
Keyword(1) | Blind Source Separation |
Keyword(2) | EM algorithm |
Keyword(3) | speech sparseness |
Keyword(4) | time-frequency masking |
1st Author's Name | Yosuke IZUMI |
1st Author's Affiliation | Graduate School of Information Science and Technology, The University of Tokyo() |
2nd Author's Name | Hirokazu KAMEOKA |
2nd Author's Affiliation | Graduate School of Information Science and Technology, The University of Tokyo |
3rd Author's Name | Nobutaka ONO |
3rd Author's Affiliation | Graduate School of Information Science and Technology, The University of Tokyo |
4th Author's Name | Shigeki SAGAYAMA |
4th Author's Affiliation | Graduate School of Information Science and Technology, The University of Tokyo |
Date | 2006-12-15 |
Paper # | EA2006-96 |
Volume (vol) | vol.106 |
Number (no) | 432 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |