Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
This presentation describes the analysis of the prosodic features of intonation and rhythm within the SPELL system, a workstation for the automated assessment and improvement of English, French and Italian pronunciation by non-native speakers. For each language, a limited range of phonologically distinctive intonation contours has been chosen. These contours are characterized using a system of pitch anchor points and pitch trajectories. A similarity metric evaluates the acceptability of a student's intonation using a smoothed fundamental frequency contour and an automatic segmentation of the student's utterance derived by a Hidden Markov Model (HMM) technique. The analysis of rhythm concentrates on the control of salience relationships within an utterance (the contrast between weak and strong syllables) using the parameters of vowel quality and duration only. Judgements arc expressed in terms of the weak-strong syllable contrast, obtained indirectly using the HMM segmenter with phrase models which allow for errors on these two parameters.
Bibliographic reference. Rooney, Edmund / Hiller, Steven M. / Laver, John / Jack, Mervyn A. (1992): "Prosodic features for automated pronunciation improvement in the spell system", In ICSLP-1992, 413-416.