| Abstract #717, Date 2/17/99, Session U, Podium , 4:30p |
| Performance of a simple feedback model of F0 control |
| *T.C. Hain, T.A. Burnett, C.R. Larson (Northwestern University) |
Control of voice F0 is primarily open-loop but there is also compelling evidence for sensory feedback, apparently used for fine-tuning. Recently there has been a rapid increase the amount of experimental data regarding feedback control. Recent studies using the pitch-shifting technique (PSS) have demonstrated when voice pitch feedback is unexpectedly shifted, people respond with a change in voice F0 output that has a latency of 100 - 150 ms (Burnett, et al, 1998; Larson, 1994). Work in progress by our group also documents feedback related responses to gradually changing voice pitch feedback (ramp stimuli), as well as to delayed auditory feedback (DAF). In an attempt to organize this extensive body of experimental data and to place it within the context of our present understanding of neuroanatomy and physiology, we have developed a mathematical model of feedback control of F0. We have explicitly attempted to build a model consistent with known anatomic pathways, in the hopes that it may be a guide for future experimental work. There are several critical and unusual features to auditory feedback in the control of F0. While feedback processes generally involve comparison of a sensory signal with a reference, in F0 control the reference signal may be derived internally from memory, externally from auditory input, or from a blend of internal and external references. The sign of the feedback signal can vary depending on the reference source. The sensory feedback is probably mainly derived from auditory input. A second problem is that there are unavoidable biological delays. In feedback loops, delays can lead to instability. The third problem is that the system must be highly accurate. These problems were solved, demonstrating feasibility, by controlling F0 with an error signal constructed by subtracting expected, internally delayed F0 from actual F0, which is subject to production and registration delays. Low pass filtering was necessary to match dynamics, and a limiter was also necessary to duplicate saturation behavior. This simple model, though by no means complete, should provide a basis for future experiments. |