To facilitate the access to, and analysis of, the large body of speech articulation data being accumulated with an x‐ray microbeam system [Fujimura, Miller, and Kiritani, J. Acoust. Soc. Am. 60, S64 (1976)], pattern‐matching procedures have been developed to link these data bases with phonetic transcriptions of the speech material. A method previously used for isolated‐word data [Nelson, Miller, and Macchi, J. Acoust. Soc. Am. 63, S32 (1978)] is now extended to process the continuous speech data. The sequence of stressed vowels and nonvoiced consonants in the phonetic transcription are matched with occurrences of mandible‐height minima and nonvoiced segments in the articulatory and acoustic data. The minimum‐cost solution for matching these two types of events is found by a dynamic programming procedure, using a cost function based on the overall error in predicting the duration of these events and the intervals between them, plus a cost for the nonoccurrence of predicted articulation events in these intervals.

This content is only available via PDF.