Toward a model of compensation

Tim Mills

My paper will present literature bearing on perturbed speech and what it can tell us about speech production. I will also present a plan of experiments that will be conducted to further explore perturbed speech and its implications for speech production mechanisms.

In examining unperturbed speech, we find that a large number of articulatory and acoustic patterns relate to any one linguistic unit of contrast (segments, features, etc). Phonological voicing contrasts, for example, are reflected in the acoustics as differences in consonant duration, differences in vowel duration, extent of vocal pulses (phonetic voicing) during the consonant, voice onset time following the consonant, and F0 and formant perturbations in adjacent vowels. If we only look at normal speech, we will always see them all present (in their respective environments), and it will be difficult to determine which are intended output of the speech production mechanism and which are accidental byproducts.

If we perturb the speech signal in some way, we can begin to tease this aspect of speech production apart. For example, if the speaker whispers instead of speaking normally, certain cues normally present in speech cannot be produced, such as voice onset time and F0. How the speaker responds to this disruption gives us information about the role of these cues in the speech production process. Will the remaining cues be increased in magnitude to compensate? Will the affected cues be encoded in a different way, or new ones introduced to maintain contrast? Or will the speaker simply ignore the disruption and speaker as if nothing is wrong, letting the overall redundancy of the acoustic signal fall?

Other perturbations I will examine include bite-block experiments, speech with an artificial source, pipe speech, and ventriloquy.