Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

Memory-Enhanced MMSE-Based Channel Error Mitigation for Distributed Speech Recognition

Cheng-Lung Lee, Wen-Whei Chang

National Chiao-Tung University, Taiwan

A new approach to sequence minimum mean-squared error (SMMSE) decoding for vector quantization over channels with memory is presented. The decoder is based on the Gilbert channel model that allows the exploitation of intra-vector correlation of bit error sequences. We apply the memory-enhanced SMMSE decoding algorithm to channel error mitigation in distributed speech recognition. Experiments on Mandarin digit string recognition task indicate that with the aid of Gilbert channel characterization, the proposed scheme obtains better performance than the ETSI mitigation algorithm under GSM channel conditions.

Full Paper

Bibliographic reference.  Lee, Cheng-Lung / Chang, Wen-Whei (2005): "Memory-enhanced MMSE-based channel error mitigation for distributed speech recognition", In INTERSPEECH-2005, 3137-3140.