Presentation | 2001/12/13 Rapid Model Adaptation with a Prior Noise GMM and Multi-SNR Models for Noisy Speech Recognition Masaki IDA, Satoshi NAKAMURA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | When a speech recognition system is used in a real environment, the recognition performance is affected by surrounding noise. Most additional noises are difficult to predict about kind of noise and SNR, so we cannot avoid the mismatch situation between those of training data and test data. Then we need a method to deal with mismatched noise problems and unknown SNRs. In this paper, we propose an HMM composition-based model adaptation that uses a prior noise data against noise mismatches. We also prepare plural HMMs for several SNRs and select the best model based on acoustic likelihood to deal with the unknown SNRs. Experimental results with AURORA2 task show 53% word accuracy improvement from baseline system with 1 sec real noise data for adaptation. The performance is equivalent to a case with 10 sec real data using the conventional HMM composition method. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | HMM composition / noise model / nonstationary noise / multipath model |
Paper # | NLC2001-57,SP2001-92 |
Date of Issue |
Conference Information | |
Committee | SP |
---|---|
Conference Date | 2001/12/13(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) | Rapid Model Adaptation with a Prior Noise GMM and Multi-SNR Models for Noisy Speech Recognition |
Sub Title (in English) | |
Keyword(1) | HMM composition |
Keyword(2) | noise model |
Keyword(3) | nonstationary noise |
Keyword(4) | multipath model |
1st Author's Name | Masaki IDA |
1st Author's Affiliation | ATR Spoken Language Translation Research Laboratories() |
2nd Author's Name | Satoshi NAKAMURA |
2nd Author's Affiliation | ATR Spoken Language Translation Research Laboratories |
Date | 2001/12/13 |
Paper # | NLC2001-57,SP2001-92 |
Volume (vol) | vol.101 |
Number (no) | 522 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |