Syllable Network for Phonemic Decoding of Speech

Authors

  • V. Gupta
  • M. Lennig
  • J. Marcus
  • P. Mermelsein

Abstract

The decoding of speech into phonemes for large vocabulary speech recognition is made more reliable by restricting phoneme sequences to those which compose valid syllables. To apply this restriction when decoding a sequence of phonemes, we use a syllable network representing the valid syllables in Webster’s 7th Collegiate dictionary. Since major allophonic variants of a phoneme are determined by the phoneme's position within the syllable (e.g., prevocalic vs. postvocalic /r /), the syllable network can be used to represent allophonic variation by employing distinct allophone models of a phoneme in different positions within the network. A preliminary experiment using the syllable network in large vocabulary recognition to select appropriate Markov models for allophones shows promising results.

Additional Files

Published

2022-12-03

How to Cite

1.
Gupta V, Lennig M, Marcus J, Mermelsein P. Syllable Network for Phonemic Decoding of Speech. Canadian Acoustics [Internet]. 2022 Dec. 3 [cited 2024 Nov. 23];14(3 bis):45-6. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3515

Issue

Section

Proceedings of the Acoustics Week in Canada