Abstract:
Variability in speech signals for the same linguistic message, both systematic and random, can be described by an abstract representation of an utterance as manipulation of control variables. Random variation of linguistic variables are often considered speech errors. The C/D model converts segmental and prosodic phonological specifications supplemented by utterance parameters such as speed and style of utterance, speaker's anatomical and physiological characteristics, emotional state, etc. into continuously variable signals. Such parameter manipulations involve slower changes in system parameters including coefficients for relating the syllable magnitude to syllable component durations, certain limited modifications of parameters in impulse response function (IRF) specifications for elemental consonantal gestures, etc. The relative stability of IRF's provides a solid basis for the perceptual strategy in segmental identification, whereas relatively free control of base functions related to laryngeal, mandibular, and vocalic dorsal/labial adjustments, possibly with auditory feedback, constitute prosodic variability of the message. Contributions of prosodic variability to the languages' historical evolution and child language development are also noted. [Work supported in part by research and education funds provided by ATR/ITL and ATR/HIP.]