Integration of speech enhancement and coding techniques

Speech coding techniques commonly used in low bit rate analysis-by-synthesis linear predictive coders (LPAS coders) can serve as a speech signal model emphasizing its important features. In the paper it is shown how this coding method can be utilized for speech enhancement. Particularly, the speech signal is modeled as the output of a cascade of an adaptive formant filter and a pitch filter, driven by a white Gaussian process with variance changing with time. A signal estimation method based on the Kalman filter is investigated which implements this speech signal model. The proposed approach yields significantly better performance both in SNR and subjective impression than Kalman filter methods, which use only short-time speech parameters.