| Presentation | 1994/10/13 Unified speaker adaptation algorithm of maximum a posteriori estimation and transfer vector field smoothing Masahiro Tonomura, Tetsuo Kosaka, Shoichi Matsunaga, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | A noise processing is essential for a real world speech recognizer.A spectral subtraction(SS)is easy to introduce into an HMM base speech recognizer.In the SS,better performance can be obtained by choosing the ratio for noise subtraction according to the SNR.In this paper,we investigate an estimated segmental SNR base adaptive spectral subtraction(SASS)approach for 1)high performance independent of SNR and 2)time-synchronous algorithm, etc.To introduce SASS into our system,a way of the segmental SNR estimation,a ratio function for noise subtraction according to the segmental SNR and the usage of of differential parameter are studied.We evaluate SASS using our speaker independent,large vocabulary,continuous speech recognizer.The results show sentence recognition rates of 80.8%,76.8%,28.0% when the SNR are 30dB,20dB, 10dB,respectively.These performance are equivalent to those of the best sentence recognition rate achieved by the ordinary SS whose ratio for noise subtraction according to SNR is a constant value. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | speech recognition / adaptive spectral subtraction / segmental SNR / time-synchronous algorithm |
| Paper # | SP94-51 |
| Date of Issue |
| Conference Information | |
| Committee | SP |
|---|---|
| Conference Date | 1994/10/13(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 | Speech (SP) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Unified speaker adaptation algorithm of maximum a posteriori estimation and transfer vector field smoothing |
| Sub Title (in English) | |
| Keyword(1) | speech recognition |
| Keyword(2) | adaptive spectral subtraction |
| Keyword(3) | segmental SNR |
| Keyword(4) | time-synchronous algorithm |
| 1st Author's Name | Masahiro Tonomura |
| 1st Author's Affiliation | Advanced Telecommunication Research Institute International() |
| 2nd Author's Name | Tetsuo Kosaka |
| 2nd Author's Affiliation | Advanced Telecommunication Research Institute International |
| 3rd Author's Name | Shoichi Matsunaga |
| 3rd Author's Affiliation | Advanced Telecommunication Research Institute International |
| Date | 1994/10/13 |
| Paper # | SP94-51 |
| Volume (vol) | vol.94 |
| Number (no) | 271 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |