RESTORATION OF MISSING VOICED SPEECH SIGNAL SEGMENTS

摘要

The new method for interpolation of intensity and period of fundamental frequency characteristics of speech signals is extensively tested when speech signal is distorted by lengthy impulsive noise or gap. The suitability of presented restoration technique is judged by evaluating quality of restored speech signal. The quality is estimated using signal to noise ratio as an objective measure and for perceptual evaluation of speech quality (PESQ) algorithm that is based on ITU recommendation P.862 was employed. It is shown that proposed method is more accurate in restoration of distorted or lost speech segment where characteristics of speech signal such as period of fundamental frequency and intensity have non-linear change in time.