Speech synthesis is part of the everyday life of many people with severe visual disabilities. For those who are reliant on assistive speech technology the possibility to choose a fast speaking rate is reported to be essential. But also expressive speech synthesis and other spoken language interfaces may require an integration of fast speech. Architectures like formant or diphone synthesis are able to produce synthetic speech at fast speech rates, but the generated speech does not sound very natural. Unit selection synthesis systems, however, are capable of delivering more natural output. Nevertheless, fast speech has not been adequately implemented into such systems to date. Thus, the goal of the work presented here was to determine an optimal strategy for modeling fast speech in unit selection speech synthesis to provide potential users with a more natural sounding alternative for fast speech output.
Das Dokument ist frei verfügbar.