A modified speech analysis&synthesis system

Abstract

Speech analysis and synthesis is an important technology in the audio processing. In this paper, a modified speech analysis and synthesis system is proposed. It adopts more abundant excitation signal which include smooth part and compensation impulse. The smooth part is the envelope curve of unvoiced sounds and the compensation impulse is represented by the peaks during pitch period. Compared with the single impulse that repeats with pitch period and the pure white noise, these signals can reflect the tone changes and characters in the period of original speech more accurately. And they make it possible for modified system to synthesize high quality speech by employing less order filter. The experimental result indicates that the modified system has advance in both speech quality and intelligibility without increasing much in code rate. For most speech, it has a good effect on recovery.