8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Robust Speech Recognition to Non-Stationary Noise Based on Model-Driven Approaches

Christophe Cerisara, Irina Illina

LORIA, France

Automatic speech recognition works quite well in clean conditions, and several algorithms have already been proposed to deal with stationary noise. The next challenge consists to work with non-stationary noise. This paper studies this problem. We propose three algorithms to non-stationary noise adaptation : Static and Dynamic Optional Parallel Model Combination (OPMC) and one algorithm derived from the Missing Data framework. The combination of speech and noise is expressed in the spectral domain and different ways to estimate the non-stationary noise model are studied. The proposed algorithms are tested on a telephone database with added background music at different SNRs. The best result is obtained using dynamic OPMC.

Full Paper

Bibliographic reference.  Cerisara, Christophe / Illina, Irina (2003): "Robust speech recognition to non-stationary noise based on model-driven approaches", In EUROSPEECH-2003, 3053-3056.