Multi voice text to speech synthesis based on the instantaneous parametric voice conversion

The paper describes an approach to text-to-speech synthesis based on processing in harmonic domain. A special harmonic analysis technique is presented that provides accurate estimation of instantaneous harmonic parameters. The technique is based on narrow band filtering aligned to the fundamental frequency, which improves estimation accuracy of higher-order harmonics with rapid frequency changes. The advanced analysis ensures natural-sounding amplitude, pitch and phase matching because of the fine deterministic/ stochastic separation. Speech synthesis is carried out using parametric representation that allows applying voice conversion techniques in order to get a multi voice synthesis system with a single voice acoustic database.