ASR2000 - Automatic Speech Recognition: Challenges for the new Millenium September 18-20, 2000 Paris, France

Noise Adaptation of HMMs Using Neural Networks

Sadaoki Furui and Daisuke Itoh

Department of Computer Science, Tokyo Institute of Technology, Japan

This paper proposes a new method, using neural networks, of adapting phone HMMs to noise added speech. The network is designed to map clean speech HMMs to noise-adapted HMMs using inputs of clean speech phone HMMs, noise HMMs and signal-to-noise ratios (S/N). The network is trained to minimize the mean squared error between the output HMMs and the target noise-adapted HMMs. Noisy broadcast-news speech was recognized in speaker-dependent and speaker-independent network training conditions, and the trained networks were confirmed to be effective in the recognition of new speakers and under new noise and S/N conditions.

Full Paper (PDF)   Full Paper (Zipped Postscript)

Bibliographic reference.  Furui, Sadaoki / Itoh, Daisuke (2000): "Noise adaptation of HMMs using neural networks", In ASR-2000, 160-167.