Presentation | 2006/12/14 Referential Reconstruction in Complex Frequency Domain for Noise Reduction Takehiro IHARA, Kazuyuki TAKAGI, Kazuhiko OZEKI, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | This paper presents a method for extracting the speech signal from the single-channel speech signal contaminted by the noise in order to improve the performance of automatic speech recognition of the noise contaminated input signal. It is assumed that the small database of utterance by the same speaker of the input signal that differ from the input signal can be used. For the same problem, the authors presented a method in [1] that extracts frames similar to the input frames by some similarity measure from the small database, and then produces output frames by refering the similar frames. In this paper, an improved similarity measure and the production process of the outputs is reported. The main improved points are keeping the phase information of Fourier transformed frames instead of discarding it, and applying the binary mask to the frames. While the phase infomation is conventionally discarded by the process of obtaining absolute spectrum, the authors consider it as worth infomation for noise reduction. Applying the binary mask to the Fourier transformed frames has the meaning similar to removing the noise component from the signal in the time domain. For evaluation, words recognition experiments by using instrumental music and environmental noise of SNR of 0dB were performed. The correctness was approximately 58%. The judgement of voiced and unvoiced speech and silent part has not been automated yet. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | noise reduction / referential reconstruction / complex frequency / frequency mask / nearest neighbor |
Paper # | NLC2006-40,SP2006-96 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2006/12/14(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) | Referential Reconstruction in Complex Frequency Domain for Noise Reduction |
Sub Title (in English) | |
Keyword(1) | noise reduction |
Keyword(2) | referential reconstruction |
Keyword(3) | complex frequency |
Keyword(4) | frequency mask |
Keyword(5) | nearest neighbor |
1st Author's Name | Takehiro IHARA |
1st Author's Affiliation | Department of Computer Science, the University of Electro-Communications() |
2nd Author's Name | Kazuyuki TAKAGI |
2nd Author's Affiliation | Department of Computer Science, the University of Electro-Communications |
3rd Author's Name | Kazuhiko OZEKI |
3rd Author's Affiliation | Department of Computer Science, the University of Electro-Communications |
Date | 2006/12/14 |
Paper # | NLC2006-40,SP2006-96 |
Volume (vol) | vol.106 |
Number (no) | 441 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |