A synthesis method based on concatenation of demisyllables and a residual excited vocal tract model

This paper describes the back-end of a new, flexible, high-quality TTS system. Preliminary results have demonstrated a highly natural and intelligible output. Although the system follows some standard methodologies, such as concatenation, we have introduced a number of novel features and a combination of techniques that make our system unique. We will describe in detail many of the design decisions and compare them with other known systems. A demonstration of the speech quality with implanted prosody is available in waveform file ([WAVE stltts1.wav and stltts2.wav]) on the conference CD.