ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

3pID2. Hot topics in speech communication.

D. H. Whalen

Haskins Labs., 270 Crown St., New Haven, CT 06511

In speech processing, mapping from acoustics to articulation is becoming more useful. While acoustically based recognizers have achieved impressive success rates, their limitations do not seem to be yielding to further acoustic manipulations. Recovering the articulation underlying the speech offers a way of decoding the complex acoustic manifestations of simple articulatory events. These results also offer benefits in bit-rate reduction and speech synthesis. In speech production, nonlinear dynamic models of speech articulation are beginning to bear fruit. The transition from stasis to vibration of the larynx (the voicing source in speech and singing) can be viewed as a switch from a point attractor (damped oscillator) to a limit cycle (self-sustained oscillation). The tools of nonlinear mathematics and chaos theory allow us to explore such topics as voice breaks, yodeling, and certain voice disorders, as well as normal phonation, Nonlinear techniques have also provided important insights into the production mechanisms of fricative consonants. In speech perception, one of the most interesting new developments is the patterns of brain activity (seen in blood flow) shown by PET scans and by functional MRI. Researchers have found that passive listening to speech activates the temporal lobe, but making judgments on that same speech also involves Broca's area. Reading print also activates Broca's area, indicating a crucial link between speech and reading. Another finding is that males rely mostly on the left premotor cortex in performing phonetic tasks presented in print, while females are not strongly lateralized. As acquisition time declines, fMRI promises to yield even more specific information on speech perception.