Abstract

A Graph Theoretic Approach for Generating Hypotheses About Phonetic Cues in Speech

Authors
  • Anne Marie Crinnion (Harvard University)
  • Beth Malmskog (Colorado College)
  • Joe Toscano (Villanova University)

Abstract

Current models of speech perception suggest that combining acoustic cues and factoring out contextual variability allows listeners to recognize speech across different talkers. However, it remains unclear which specific cues are necessary and how their use varies between individual talkers. We use graph theoretic techniques to address these problems by constructing networks connecting talkers and possible cues. We identify subgraphs (Steiner trees) that connect talkers via cues consistently used to indicate specific phonemes. Classifiers trained on these cues match listeners\' data better than those trained on all cues, suggesting that Steiner trees can identify the cues necessary for speech recognition.

Keywords: speech perception, phonetics, graph theory

How to Cite:

Crinnion, A., Malmskog, B. & Toscano, J., (2018) “A Graph Theoretic Approach for Generating Hypotheses About Phonetic Cues in Speech”, Society for Computation in Linguistics 1(1), 211-212. doi: https://doi.org/10.7275/R5BG2M5X

Downloads:
Download PDF

53 Views

16 Downloads

Published on
01 Jan 2018