15th International Congress of Phonetic Sciences (ICPhS-15)Barcelona, Spain |
This paper presents a new approach for designing a Concatenative text-to-speech (TTS) system based on multi-domain unit selection. The method achieves good synthetic quality with reasonable computational cost for a general-purpose TTS system. The architecture of the multi-domain database and the text classification algorithm for domain assignment are the basis of the method. The performance of the adjusted text classification algorithm for the multi-domain TTS aim is analyzed in several encouraging experiments.
Bibliographic reference. Alías, Francesc / Iriondo, Ignasi / Barnola, Pere (2003): "Multi-domain text classification for unit selection text-to-speech synthesis", In ICPhS-15, 2341-2344.