next up previous
Next: Bibliography Up: Optimal Data Selection for Previous: Future Work


The generated prompt list, examples and code used to select data is distributed as part of FestVox. On line examples discussed in this paper are available from

We presented a method for finding a small set of utterances from a large database that optimally cover the acoustic-phonetic space with respect to a particular speaker. The quality of voice produced from the selection is far superior that a voice build form a more general database. The resulting voice is better for data, while retaining the style of the database from which it was trained.

Alan W Black 2001-08-26