In this paper, a single-channel speech enhancement method by coupling speech detection and spectral amplitude estimation is proposed. First, the optimal detector used for the spectral coefficients of the speech signal is obtained by minimizing the combined risk function which considers both detection and estimation errors. Second, according to the optimal speech detector, the optimal spectral amplitude estimator is obtained by further minimizing the combined risk function. For speech-presence and speech-absence, we propose two general weighted cost functions which are based on human auditory perception and compressive nonlinearities of the cochlea. The cost parameters are employed to balance the speech distortion and residual noise caused by missed detection and false alarm, respectively. The performance of the proposed method is evaluated under ITU-T G.160. The experimental results indicate that the proposed method evidently performs better than the reference methods.
Bibliographic reference. Deng, Feng / Bao, Chang-chun / Bao, Feng (2013): "A speech enhancement method by coupling speech detection and spectral amplitude estimation", In INTERSPEECH-2013, 3234-3238.