Speech Enhancement using Statistical based and NMF approaches. MMSE Estimtors and Non-Negative Matrix Factorization Approaches for Speech Enhancement

      Аннотация: The world all living around is noisy. The speech communicated from one person to other person or person to machine is contaminated by noise present in the environment. In modern speech communication systems like smart phone, teleconferencing, hearing aids and human-machine interfaces, the speech is recorded with single microphone or multiple microphones. The microphones record the speech signal of interest along with the noise that is present in environment. Due to the noise the signal quality and intelligibility reduces and it is hard to understand the speech signal clearly. Thus it is necessary to clean the speech signal for better understanding. The process of reducing noise / improving speech quality of the contaminated signal is referred to as speech enhancement. Traditional speech enhancement algorithms like spectral subtraction, Wiener filtering and MMSE estimators, process the noisy speech spectral amplitudes to improve the quality and intelligibility of the speech. Wiener filter gain is obtained from a priori SNR. But the calculation of a priori SNR depends on previous frame. So a two step noise reduction technique is developed to obtain a priori SNR from current frame.

Ключевые слова: Speech Enhancement, noise reduction, Bayesian Estimators