Multi-microphone signal subspace speech enhancement based on tensor-preprocessing

This paper proposes a tensor-preprocessing multi-microphone signal subspace approach for speech enhancement. The approach includes two parts to eliminate the noise in multi-microphone system step by step including tensor part and subspace part. Noise is preliminarily reduced in part by finding the lower-rank approximation of a three-order tensor constructed from the multi-microphone signal with tucker model in tensor part. Speech enhancement is finished by a linear filter estimated from the data covariance matrix and the estimated noise variance in subspace part. The performance of the proposed approach is evaluated with objective indexes and listening quality test. The experimental results indicate that the proposed approach has good performance of retrieving the target signal from noisy multi-microphone signal.