Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

Non-Linear Annotation of Multi-Channel Speech Data

Alain Marchal (1), William J. Hardcastle (2), K. Nicolaidis (2), N. Nguyen (1), F. Gibbon (2)

(1) Laboratoire Parole et Langage, URA 261 CNRS, Universite de Provence, Aix-en-Provence, France
(2) Speech Research Laboratory, Department of Linguistics, University of Reading, Reading, UK

ESPRIT II Basic Research Action ACCOR aims at investigating the articulatory-acoustic correlations in coarticulatory processes in seven European languages (English, French, German, Italian, Irish Gaelic, Catalan and Swedish). In order to allow for cross-language comparisons, it is particularly important that partners in the project adopt a common methodology, i.e. standardised investigation tools and normalised measurement procedures at specific locations in the speech signals. The physiologic and acoustic database first needs to be segmented and labelled according to principles that do not preclude any theoretical interpretation. The conventional phonological approach to identifying segmental boundaries that implicitly considers coarticulation as a phonological feature-spreading process is rejected. On the contrary, a non-linear annotation of the articulatory and acoustic events has been adopted, based on the evidence provided by the different channels of information: acoustic waveform, airflow traces, linguo-palatal contact patterns, jaw and lip movements. We show with reference to the analysis of /kl/ clusters how this truly multi-level approach potentially leads to a better understanding of articulatory-acoustic mapping.

Full Paper

Bibliographic reference.  Marchal, Alain / Hardcastle, William J. / Nicolaidis, K. / Nguyen, N. / Gibbon, F. (1992): "Non-linear annotation of multi-channel speech data", In ICSLP-1992, 787-790.