The manual preparation of diphone libraries is time-consuming. This paper describes a system for automatic generation of diphone libraries. In this system the incoming utterance is first split up into stationary parts , then these stationary parts are labeled as phonemes using the phonetic transcription, and subsequently the diphone boundaries are calculated from the phoneme boundaries . With this system, diphone libraries for Dutch and German have been made in only a fraction of the time that a manual method would have taken.
|Journal||IPO Annual Progress Report|
|Publication status||Published - 1985|