Ninth International Conference on Spoken Language Processing

Pittsburgh, PA, USA
September 17-21, 2006

On the Sufficiency of Automatic Phonetic Transcriptions for Pronunciation Variation Research

Christophe Van Bael, Hans van Halteren

Radboud Universiteit Nijmegen, The Netherlands

We investigated whether automatic phonetic transcriptions (APTs) can replace manually verified phonetic transcriptions (MPTs) in a large corpus-based study on pronunciation variation. To this end, we compared the performance of both transcription types in a classification experiment aimed at establishing the direct influence of a particular situational setting on pronunciation variation. We trained classifiers on the speech processes extracted from the alignments of an APT and an MPT with a canonical transcription. We tested whether the classifiers were equally good at verifying whether unknown transcriptions represent read speech or telephone dialogues, and whether the same speech processes were identified to distinguish between transcriptions of the two situational settings. Our results not only show that similar distinguishing speech processes were identified; our APT-based classifier yielded better classification accuracy than the MPT-based classifier whilst using fewer classification features.

Full Paper

Bibliographic reference.  Bael, Christophe Van / Halteren, Hans van (2006): "On the sufficiency of automatic phonetic transcriptions for pronunciation variation research", In INTERSPEECH-2006, paper 1265-Mon3FoP.7.