Multichannel speech recognition using distributed microphone signal fusion strategies

Multichannel fusion strategies are presented for the distributed microphone recognition environment, for the task of song-type recognition in a multichannel songbird dataset. The signals are first fused together based on various heuristics, including their amplitudes, variances, physical distance, or squared distance, before passing the enhanced single-channel signal into the speech recognition system. The intensity-weighted fusion strategy achieved the highest overall recognition accuracy of 94.4%. By combining the noisy distributed microphone signals in an intelligent way that is proportional to the information contained in the signals, speech recognition systems can achieve higher recognition accuracies.