QUALITY OF SPEECH PRODUCED BY ANALYSIS-SYNTHESIS

We review factors that have affected the synthesis of high-quality speech by analysis-synthesis. The influence of a selected subset of these factors on the quality of synthesized speech was evaluated through listener preference judgements by comparing natural speech to the synthetic speech of two synthesizers: linear prediction coding (LPC) and formant. Several synthesizer excitation waveforms were considered. These waveforms included critical parameters that replicated selected glottal timing events, e.g., the instants of glottal closure and glottal opening. In addition, identifying voiced/unvoiced/mixed excitation and silent intervals in the speech waveform and measuring the fundamental frequency of voicing contributed to the synthesis of high-quality speech. A two-channel approach to speech analysis is recommended to aid the automatic processing of speech, where one channel is the conventional acoustic signal, while the other channel is the electroglottogram (EGG). © 1990.