Stimulus parameters in speech compression.

Abstract

A total of 36 stimulus tapes were prepared by the conjoint manipulation of three parameters potentially critical in their control of compression effects. These parameters were: (1) talker rate, manipulated through the limits attainable by a trained speaker; (2) selective pause compression at three values, 100, 50, and 0% deletions and (3) random deletions at four values, 0, 30, 50, and 70%. Each of the 36 stimulus tapes was analyzed with respect to distribution of pause-to-phonation time and rate. In addition, the message was analyzed in terms of the Information Theory in an attempt to specify the moment-to-moment textual constraints present. Comprehension data provided by 135 Ss indicated that pause distribution in a connected message interacts non-additively with rate. Pause-to-phonation ratios in normally accelerated speech decrease logarithmically as speech rate increases and artificial treatment of these ratios produces significant over-all rate increases which survive listening tests better than some random compression treatments. Methodological considerations are discussed in some detail in an attempt to suggest certain normative measures which would make it possible to more readily compare compression effects across differing messages, speakers and treatments, and as a consequence increase the generality of research findings.