잡음음성 음향모델 적응에 기반한 잡음에 강인한 음성인식
Noise Robust Speech Recognition Based on Noisy Speech Acoustic Model Adaptation
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
In the Vector Taylor Series (VTS)-based noisy speech recognition methods, Hidden Markov Models (HMM) are usuallytrained with clean speech. However, better performance is expected by training the HMM with noisy speech. In a previousstudy, we could find that Minimum Mean Square Error (MMSE) estimation of the training noisy speech in the log-spectrumdomain produce improved recognition results, but since the proposed algorithm was done in the log-spectrum domain, it couldnot be used for the HMM adaptation. In this paper, we modify the previous algorithm to derive a novel mathematical relationbetween test and training noisy speech in the cepstrum domain and the mean and covariance of the Multi-condition TRaining(MTR) trained noisy speech HMM are adapted. In the noisy speech recognition experiments on the Aurora 2 database, theproposed method produced 10.6% of relative improvement in Word Error Rates (WERs) over the MTR method while theprevious MMSE estimation of the training noisy speech produced 4.3% of relative improvement, which shows the superiorityof the proposed method.
Abstract
In the Vector Taylor Series (VTS)-based noisy speech recognition methods, Hidden Markov Models (HMM) are usuallytrained with clean speech. However, better performance is expected by training the HMM with noisy speech. In a previousstudy, we could find that Minimum Mean Square Error (MMSE) estimation of the training noisy speech in the log-spectrumdomain produce improved recognition results, but since the proposed algorithm was done in the log-spectrum domain, it couldnot be used for the HMM adaptation. In this paper, we modify the previous algorithm to derive a novel mathematical relationbetween test and training noisy speech in the cepstrum domain and the mean and covariance of the Multi-condition TRaining(MTR) trained noisy speech HMM are adapted. In the noisy speech recognition experiments on the Aurora 2 database, theproposed method produced 10.6% of relative improvement in Word Error Rates (WERs) over the MTR method while theprevious MMSE estimation of the training noisy speech produced 4.3% of relative improvement, which shows the superiorityof the proposed method.
참고문헌 (10)
- Gales, M., "Model based techniques for noise- robust speech recognition", Ph.D., pp.1995
- Ball, S. F, "Suppression of acoustic noise in speech using spectral subtraction", IEEE Trans. Acoust., Speech, Signal Process, 27(2), pp.113~122 (1979)
- Moreno, P. J., "Speech recognition in noisy environments", Ph.D., pp.1996
- Hirsch, H. G., "The Aurora experimental framework for the performance evaluation of speech recognition systems under noisy conditions", in Proc. ICSLP, pp.18~20 (2000)
- Xu, H., "Noise condition-dependent training based on noise classification and SNR estimation", IEEE Trans. Audio, Speech and Language Process, 15(8), pp.2431~2443 (2007)
- Kalinli, O., "Noise adaptive training for robust automatic speech recognition", IEEE Trans. Audio, Speech and Language Process, 18(8), pp.1889~1901 (2010)
- Chung, Y., "Compensation of SNR and noise type mismatch using an environmental sniffing based speech recognition solution", EURASIP Journal on Audio, Speech, and Music Processing, 12, pp.1~14 (2013)
- Gopinath, R. A., "Robust speech recognition in Noise : Performance of the IBM continuous speech recognizer on the ARPA noise spoke task", in Proc. ARPA Spoken Language System Technology, pp.127~130 (1995)
- ETSI draft standard doc., Speech processing, Transmission and Quality aspects (STQ);, "Distributed speech recognition;Front-end feature extraction algorithm; Compression algorithm", ETSI Standard ES 202 050, ~ (2002)
- Hidden Markov Model Toolkit V3.4.1, "Young, S.", Cambridge University, Engineering Department, Speech Group, ~ (1993)