Skip to main content
Abstract

Modeling exemplar production from speech tokens with acoustic targets

Authors
  • Frédéric "Fred" Mailhot (Dialpad, Inc.)
  • Cassandra L. Jacobs (University of Buffalo)

Abstract

Temporal variability is a core obstacle to modelng exemplar-based production from real speech. Different tokens of a given word or utterance may vary widely in duration; a distance function over speech exemplars must account for this variability. We present an exemplar production mechanism that uses a distance measure. robust to temporal variation in order to meaningfully average over such tokens. We show qualitatively and quantitatively that the outputs produced are meaningful and retain sufficient phonetic detail to be identifiable.

Keywords: exemplar production, speech

How to Cite:

Mailhot, F. ". & Jacobs, C. L., (2024) “Modeling exemplar production from speech tokens with acoustic targets”, Society for Computation in Linguistics 7(1), 281–284. doi: https://doi.org/10.7275/scil.2162

Downloads:
Download PDF

229 Views

81 Downloads

Published on
2024-06-24

Peer Reviewed