Presentation 2005/6/9
A Study of Running Spectum Filtering based on the estimated SNR
Noboru HAYASAKA, Shingo YOSHIZAWA, Naoya WADA, Yoshikazu MIYANAGA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We proposed RSF and DRA as the noise robust methods. The RSF is a technique for applying the filter to the spectrum twice. This technique has the feature of speech enhancement, reduction of convolutive noise and a part of additive noise. However, the RSF method removes a part of the speech. Therefore, the recognition rates are deteriorated in clean environment. We propose the RSF based on the estimated SNR. The estimated SNR calculate each frame and frequency. As a result of the isolated word recognition experiments, the recognition rates are improved in high SNR and clean environments.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Estimated SNR / RSF / Robust / Word speech recognition
Paper # SIS2005-2
Date of Issue

Conference Information
Committee SIS
Conference Date 2005/6/9(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 Smart Info-Media Systems (SIS)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Study of Running Spectum Filtering based on the estimated SNR
Sub Title (in English)
Keyword(1) Estimated SNR
Keyword(2) RSF
Keyword(3) Robust
Keyword(4) Word speech recognition
1st Author's Name Noboru HAYASAKA
1st Author's Affiliation Graduate school of infomation science and technology()
2nd Author's Name Shingo YOSHIZAWA
2nd Author's Affiliation Graduate school of infomation science and technology
3rd Author's Name Naoya WADA
3rd Author's Affiliation Graduate school of infomation science and technology
4th Author's Name Yoshikazu MIYANAGA
4th Author's Affiliation Graduate school of infomation science and technology
Date 2005/6/9
Paper # SIS2005-2
Volume (vol) vol.105
Number (no) 111
Page pp.pp.-
#Pages 5
Date of Issue