15th International Congress of Phonetic Sciences (ICPhS-15)Barcelona, Spain |
In fricative-vowel syllables, the perceptual identification of the consonant is significantly better when at least part of the accompanying vowel is available, with respect to the case in which only the consonant is heard. In this study, the ability of hidden markov models (HMM) for reproducing the process of phonetic integration is assessed by comparing automatic recognition with listeners' perception. Results show that HMM perform better than listeners for isolated fricative noises, but worse for fricative+vowel segments. The utility of fricative+vowel models versus syllable models is discussed.
Bibliographic reference. Fernández, Santiago / Feijóo, Sergio (2003): "Comparing HMM-based recognition to perceptual phonetic integration in fricative-vowel syllables", In ICPhS-15, 1433-1436.