Presentation	2008-01-25 Noisy Speech Recognition Using Spectral Subtraction with an Adaptive SNR Method Jun TOYAMA,
PDF Download Page	PDF download Page Link
Abstract(in Japanese)	(See Japanese page)
Abstract(in English)	Spectral Subtraction method is well known the effective reduction method for corrupted speech with stationary noise. But, real noise is not always stationary. A noise reduction method dealing with variations of noise power using a optimization method is proposed. Additionally, omputational costs are improved by increasing computation that does not depend on acoustic models, when Taylor expansion is applied to irrational expressions in the criterion function. As results form the evaluation using AURORA-2J, the propose method without using Taylor expansion achieves word accuracies of 23% 77% and 97% for clean training condition under the SNR condition 0dB, 10dB and 20dB, respectively.
Keyword(in Japanese)	(See Japanese page)
Keyword(in English)	Spectral Subtraction method / Variation of Noise Power / Optimize Problem / Taylor Expansion
Paper #	TL2007-74,SP2007-169,WIT2007-74
Date of Issue

Conference Information
Committee	TL
Conference Date	2008/1/18(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	Thought and Language (TL)
Language	JPN
Title (in Japanese)	(See Japanese page)
Sub Title (in Japanese)	(See Japanese page)
Title (in English)	Noisy Speech Recognition Using Spectral Subtraction with an Adaptive SNR Method
Sub Title (in English)
Keyword(1)	Spectral Subtraction method
Keyword(2)	Variation of Noise Power
Keyword(3)	Optimize Problem
Keyword(4)	Taylor Expansion
1st Author's Name	Jun TOYAMA
1st Author's Affiliation	Graduate School of Information Science and Technology, Hokkaido University()
Date	2008-01-25
Paper #	TL2007-74,SP2007-169,WIT2007-74
Volume (vol)	vol.107
Number (no)	433
Page	pp.pp.-
#Pages	6
Date of Issue