Automatic alignment of phonetic events with x‐ray microbeam articulatory data and the acoustic speech signal

This paper describes an algorithm for the automatic alignment of phonetic events with x‐ray microbeam articulation data and the corresponding acoustic signal. The algorithm uses a two‐step procedure similar to that of Nelson [Nelson et al., J. Acoust. Soc. Am. Suppl. 1 63, S32 (1978); Nelson, AT&T Bell Laboratories Internal Rep. (1978)]. The first step locates the phrase boundary in continuous speech, and the second step matches the phonetic segments in each phrase. Articulatory and acoustic events are recognized in continuous speech, and matched to the predicted phonetic events using a dynamic programming technique. The place of articulation and voicing for certain phonemes are also matched with articulatory and acoustic events. The system is presently speaker dependent, and has to be trained to the articulatory data for the particular speaker. [Work supported by Ohio State University Speech and Heating Department.]