Aligning phonetic transcriptions with their citation forms

Abstract

One of the main motivations for publishing this paper is to make available a matrix of phone-distance measures which may be useful in dealing with large corpora of conversational speech. The paper reports how this matrix of phone-distances was created from transcriber labeling disagreements, and how it can be used in a dynamic time warping algorithm to align phonetic transcriptions of conversational speech with their citation forms. The weighted string edit distance produced by the phone-distance DTW algorithm may also be useful in calculating neighborhood densities for studies of auditory word recognition. ©2004 Acoustical Society of America.