Use this URL to cite or link to this record in EThOS:
Title: Two uses for syllables in a speech recognition system
Author: Blokland, Robert Arthur
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1994
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Many acoustic and phonetic studies attest to the value of the syllable as a unit of linguistic description. Phenomena of coarticulation and assimilation often occur within a syllable, and the acoustic realisation of some phenomes also correlate with their position in a syllable. For example, stops are more often released at the beginning of a syllable than at the end, and the clear allophone of 'I' is also more often found at the beginning than the dark allophone. Despite the prevalence of such studies, there have been relatively few attempts to apply this knowledge in a speech recognition system. This is what is attempted here. The matter is investigated by modifying the CSTR speech recogniser, which is not based on syllables. It is a modular system, with a separate front end and back end. The front end uses hidden Markov models to produce a phoneme lattice, and the back end uses a dynamic programming algorithm to construct words out of the lattice. Syllable information is incorporated in both the front and the back end. Because the system is modular, the effects of incorporating syllables can be studied independently. The experiments in the front end centre on the choice of a set of allophones that correlate well with their position in a syllable. Segmentation in then constrained to produce only those segment sequences that form valid syllables. The experiments in the back end centre on specialising the confusion matrix for syllable position.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available