November 1, 2013

Psycholinguistics Group Meeting (Nov 1)

The Psycholinguistic Group is meeting Friday, November 1st in Sid Smith 560A and will start at 10:15. Morgan Sonderegger (http://people.linguistics.mcgill.ca/~morgan/) is visiting from McGill that day and will give a talk entitled "Voice onset time: automatic measurement and corpus studies"

"Voice onset time: automatic measurement and corpus studies"
 Large corpora of speech from laboratory and naturalistic settings are becoming increasingly available and easy to construct, and promise to change the questions researchers can ask about human speech production. However, this promise depends on the development of accurate algorithms to quicken or replace manual measurement, which becomes infeasible for large corpora. With some important exceptions (e.g. vowel formants), such algorithms do not currently exist for most quantities which are widely measured in phonetic research. The first part of this talk describes an automatic measurement algorithm for perhaps the most widely measured consonantal variable, voice onset time (VOT), which has been extensively studied since the 1960s (Lisker & Abramson, 1964). Our approach combines knowledge about the cues human annotators use to measure VOT with machine learning techniques for predicting structured output, to tailor an algorithm which learns to measure VOT nearly as accurately as humans (evaluated on several corpora), by training on a small number (50-200) of hand-labeled examples.

The second part of the talk will describe a corpus study of variability in VOT in British reality television speech enabled by our automatic measurement algorithm, which quantifies the relative importance of different factors affecting VOT in conversational speech (e.g. speaking rate, place of articulation, speaker gender). There have been very few previous studies of VOT in conversational speech (Yao, 2009); in addition to replicating many findings of laboratory studies, we find a number of novel and surprising effects, including inter-speaker differences in the strength of conditioning factors. Time permitting, we will also describe a a collaborative project applying the algorithm to studying real-time change in VOT in a speech community.

No comments:

Post a Comment