Presentation | 2015-03-02 Optimization of impulse responses for model training in reverberant speech recognition Takahiro FUKUMORI, Masato NAKAYAMA, Takanobu NISHIURA, Yoichi YAMASHITA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | The reverberant speech degrades the speech recognition performance in the field of distant-talking speech. As one of approaches against the reverberation, a method has been studied by constructing an acoustic model trained with the reverberant speech. There is, however, a problem that the reverberation time, which is an unique value in a room, has a limitation for an improvement of speech recognition performance because speech recognition performance greatly depends on the arrangement of a speaker and a microphone. To overcome this problem, we have already proposed a construction method of an optimum acoustic model based on D value. In this method, an acoustic model is constructed by using many impulse responses measured under the conditions with different D value. The proposed method could improve speech recognition performance in comparison with a conventional method. In this paper, we try to optimize impulse responses for training an acoustic model in terms of a border time of D value and a number of impulse responses. In evaluation experiments, we confirmed higher speech recognition performance by using nine impulse responses with different reverberation time and D_<50>. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Speech recognition / Reverberation / Acoustic model / Acoustic parameters |
Paper # | EA2014-78,SIP2014-119,SP2014-141 |
Date of Issue |
Conference Information | |
Committee | EA |
---|---|
Conference Date | 2015/2/23(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) | Optimization of impulse responses for model training in reverberant speech recognition |
Sub Title (in English) | |
Keyword(1) | Speech recognition |
Keyword(2) | Reverberation |
Keyword(3) | Acoustic model |
Keyword(4) | Acoustic parameters |
1st Author's Name | Takahiro FUKUMORI |
1st Author's Affiliation | Graduate School of Information Science and Engineering, Ritsumeikan University() |
2nd Author's Name | Masato NAKAYAMA |
2nd Author's Affiliation | College of Information Science and Engineering, Ritsumeikan University |
3rd Author's Name | Takanobu NISHIURA |
3rd Author's Affiliation | College of Information Science and Engineering, Ritsumeikan University |
4th Author's Name | Yoichi YAMASHITA |
4th Author's Affiliation | College of Information Science and Engineering, Ritsumeikan University |
Date | 2015-03-02 |
Paper # | EA2014-78,SIP2014-119,SP2014-141 |
Volume (vol) | vol.114 |
Number (no) | 473 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |