Subspace-Based Multi-Channel Speech Enhancement Using a Novel Signal Subspace Dimension Estimator in Reverberant Environments

Although the SSA has been studied extensively for speech enhancement, not too much attention has been paid to discuss the method to identify signal subspace dimension. In this paper we present a novel signal subspace dimension estimator based on Frobenius norm, with which subspace-based multi-channel speech enhancement is robust to such adverse acoustic environments as room reverberation and low input SNR. Experiments show the presented method leads to more noise reduction and less speech distortion comparing to traditional methods.