A Divide and Conquer Strategy for Musical Noise-free Speech Enhancement in Adverse Environments.

A divide and conquer strategy for enhancement of noisy speeches in adverseenvironments involving lower levels of SNR is presented in this paper, wherethe total system of speech enhancement is divided into two separate steps. Thefirst step is based on noise compensation on short time magnitude and thesecond step is based on phase compensation. The magnitude spectrum iscompensated based on a modified spectral subtraction method where thecross-terms containing spectra of noise and clean speech are taken intoconsideration, which are neglected in the traditional spectral subtractionmethods. By employing the modified magnitude and unchanged phase, a procedureis formulated to compensate the overestimation or underestimation of noise byphase compensation method based on the probability of speech presence. Amodified complex spectrum based on these two steps are obtained to synthesize amusical noise free enhanced speech. Extensive simulations are carried out usingthe speech files available in the NOIZEUS database in order to evaluate theperformance of the proposed method. It is shown in terms of the objectivemeasures, spectrogram analysis and formal subjective listening tests that theproposed method consistently outperforms some of the state-of-the-art methodsof speech enhancement for noisy speech corrupted by street or babble noise atvery low as well as medium levels of SNR.