A pattern-matching technique for compressed digital voice communication was investigated. The method, based on the use of vocoders for input- and output-speech processing, used digital spectrum pattern matching for real-time compression and expansion of speech. From analysis of speech samples, catalogs of voice patterns were compiled together with data on the probabilities of occurrence of the voiced and unvoiced spectrum patterns comprising the distributions, and the average intensity level at which patterns occurred, as functions of the fidelity criterion used in classifying patterns. A strategy was established for constructing tables of pattern stereotypes, in which the fidelity criterion assigned to pattern classes was ordered as a function of pattern probability, such that frequent patterns would be matched and encoded with high precision, infrequent patterns with a larger tolerable input/output error, on a 7-valued fidelity scale. The pattern tables assembled in this manner were utilized for real-time pattern matching, in which voice samples were analyzed, categorized in terms of a “best match” among the pattern stereotypes, and synthesized from the “standardized” patterns. Speech-data items were encoded as independent samples at 50 items per second; the information-data rate was 250 bits per second for pitch data (5 bits per sample), plus 50 (n) bits per second for pattern data, where 2n = Np (number of pattern items comprising a catalog). Input/output pattern error distributions were compiled to test the efficacy of this strategy in minimizing the average errors between input and output patterns for a given catalog size for pattern matching.