An adaptive subspace approach for speech enhancement

A new algorithm is proposed for enhancement of speech degraded by colored additive interference. It is observed that the noise energy is disparately distributed along each eigenvector. These energies are obtained from noise samples gathered from silence intervals between speech samples. To obtain these silence intervals, we proposed an efficient voice activity detector based on outputs of principle component eigenfilter. Enhancement is performed by modifying each component due to its noise and clean speech energies. Tests indicated that 71% of our listeners preferred the enhanced speech by the above method over former methods for white Gaussian noise, while 80% of them preferred our method when we processed real noisy speech gathered from various environments.