Presentation | 2017-10-19 Speech enhancement of utterance while playing with werewolf game "JINRO" based on NMF Shunsuke Kawano, Toru Takahashi, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We describe that speech enhancement for natural and multi speaker dialognue. To record natural and multi speaker dialognue, we focus on utterances in the werewolf game JINRO which is one of the speech communication based card game. As we think that player’s utterance while they play with JINRO is natural, we record their dialognue and analyise it. We propose NMF based speech enhancement method to enhance one of the speakers. AfterselectingspeechsegmentofthetargetspeakerandanalysingbyNMF,his/hercharactoristicrepresentsasbase vectors. Multi speaker dialognue is analyised by NMF with the speaker’s base vectors and new base vectors. After all, noise’s and other speaker’s charactoristics are estimated in the new base vectors. The idea is that the given the speaker’s base vectors are ?xed in matrix factorization procedure. We can get activation matrix in which each line is divided into target speaker and the others. By using target speakers activities, we can synthesize enhanced speech. We conducted an experiment by 7 players, but the proposed method did not show good performance. The reason is that it is hard to get only the target speaker’s base vectors. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Jinro / Mixed speech / Sound Source Separation / Nonnegative Matrix Factorization |
Paper # | SP2017-35,WIT2017-31 |
Date of Issue | 2017-10-12 (SP, WIT) |
Conference Information | |
Committee | WIT / SP |
---|---|
Conference Date | 2017/10/19(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Tobata Library of Kyutech (Kitakyushu) |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Chikamune Wada(Kyushu Inst. of Tech.) / Yoichi Yamashita(Ritsumeikan Univ.) |
Vice Chair | Daisuke Wakatsuki(Tsukuba Univ. of Tech.) / Hiroki Mori(Utsunomiya Univ.) |
Secretary | Daisuke Wakatsuki(Nagoya Inst. of Tech.) / Hiroki Mori(AIST) |
Assistant | Takeaki Shionome(*) / Manabi Miyagi(Tsukuba Univ. of Tech.) / Takashi Handa(Saitama Industrial Technology Center) / Kei Hashimoto(Nagoya Inst. of Tech.) / Satoshi Kobashikawa(NTT) |
Paper Information | |
Registration To | Technical Committee on Well-being Information Technology / Technical Committee on Speech |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Speech enhancement of utterance while playing with werewolf game "JINRO" based on NMF |
Sub Title (in English) | |
Keyword(1) | Jinro |
Keyword(2) | Mixed speech |
Keyword(3) | Sound Source Separation |
Keyword(4) | Nonnegative Matrix Factorization |
1st Author's Name | Shunsuke Kawano |
1st Author's Affiliation | Osaka Sangyo University(OSU) |
2nd Author's Name | Toru Takahashi |
2nd Author's Affiliation | Osaka Sangyo University(OSU) |
Date | 2017-10-19 |
Paper # | SP2017-35,WIT2017-31 |
Volume (vol) | vol.117 |
Number (no) | SP-250,WIT-251 |
Page | pp.pp.7-12(SP), pp.7-12(WIT), |
#Pages | 6 |
Date of Issue | 2017-10-12 (SP, WIT) |