7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Evaluation of Noisy Speech Recognition Based on Noise Reduction and Acoustic Model Adaptation on the Aurora2 Tasks

M. Fujimoto, Yasuo Ariki

Ryukoku University, Japan

In this paper, we have evaluated a noisy speech recognition method based on noise reduction and acoustic model adaptation, on the AURORA2 tasks.

For noise reduction method, we employed two noise reduction methods. One is an Adaptive Sub-Band Spectral Subtraction (ASBSS) method which can optimize the noise subtraction rate according to the SNR in frequency bands at each frame. The other is a Kalman filtering estimation method which re-estimates the accurate speech spectra from those estimated by ASBSS. The accurate speech spectra was estimated by combining these methods. Usually, a noise reduction method has a problem that it degrades the recognition rate because of spectral distortion caused by residual noise occurred through noise reduction and over estimation. To solve the problem in noise reduction method, adaptation of the acoustic models is employed by using an unsupervised MLLR adaptation to the spectral distortion.

In evaluation on the AURORA2 tasks, our method showed the significant improvement in recognition accuracy for both clean training condition and multi training condition.

Full Paper

Bibliographic reference.  Fujimoto, M. / Ariki, Yasuo (2002): "Evaluation of noisy speech recognition based on noise reduction and acoustic model adaptation on the Aurora2 tasks", In ICSLP-2002, 465-468.