Presentation 2008-12-10
Study on Spectro-Temporal Features Based on Gradient Histograms
Takashi MUROI, Tetsuya TAKIGUCHI, Yasuo ARIKI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper proposes a novel feature extraction method for speech recognition based on gradient features on 2-D time-frequency matrix. Widely used MFCC features lack temporal dynamics and delta-MFCC is an indirect expression of temporal frequency changes. To extract the temporal dynamics more directly, local gradient features are measured in the region around reference positions. This method was originally proposed as HOG (Histograms of Oriented Gradients) and applied to human body detection in image recognition. In this paper, we develop it into gradient-based acoustic features in speech recognition. The proposed feature was evaluated on a phoneme recognition task and showed the significant improvement for clean speech and even for the noisy speech when combined with MFCC.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) gradient histograms / spectro-temporal features / phoneme recognition
Paper # NLC2008-51,SP2008-106
Date of Issue

Conference Information
Committee NLC
Conference Date 2008/12/2(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 Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Study on Spectro-Temporal Features Based on Gradient Histograms
Sub Title (in English)
Keyword(1) gradient histograms
Keyword(2) spectro-temporal features
Keyword(3) phoneme recognition
1st Author's Name Takashi MUROI
1st Author's Affiliation Graduate School of Engineering, Kobe University()
2nd Author's Name Tetsuya TAKIGUCHI
2nd Author's Affiliation Graduate School of Engineering, Kobe University
3rd Author's Name Yasuo ARIKI
3rd Author's Affiliation Graduate School of Engineering, Kobe University
Date 2008-12-10
Paper # NLC2008-51,SP2008-106
Volume (vol) vol.108
Number (no) 337
Page pp.pp.-
#Pages 5
Date of Issue