Richard Held

Adapting to supernormal auditory localization cues. I. Bias and resolution

Head-related transfer functions (HRTFs) were used to create spatialized stimuli for presentation through earphones. Subjects performed forced-choice, identification tests during which allowed response directions were indicated visually. In each experimental session, subjects were first presented with auditory stimuli in which the stimulus HRTFs corresponded to the allowed response directions. The correspondence between the HRTFs used to generate the stimuli and the directions was then changed so that response directions no longer corresponded to the HRTFs in the natural way. Feedback was used to train subjects as to which spatial cues corresponded to which of the allowed responses. Finally, the normal correspondence between direction and HRTFs was reinstated. This basic experimental paradigm was used to explore the effects of the type of feedback provided, the complexity of the stimulated acoustic scene, the number of allowed response positions, and the magnitude of the HRTF transformation subjects had to learn. Data showed that (1) although subjects may not adapt completely to a new relationship between physical stimuli and direction, response bias decreases substantially with training, and (2) the ability to resolve different HRTFs depends both on the stimuli presented and on the state of adaptation of the subject.

ADAPTING TO SUPERNORMAL AUDITORY LOCALIZATION CUES. II. CONSTRAINTS ON ADAPTATION OF MEAN RESPONSE

A series of experiments was performed in which subjects were trained to interpret auditory localization cues arising from locations different from their normal spatial positions. The exact pattern of mean response to these alterations (as a function of time) was examined in order to begin to develop a quantitative model of adaptation. Mean responses were roughly proportional to the normal position associated with the localization cues presented. As subjects adapted, the best-fit slope (relating mean response and normal position) changed roughly exponentially with time. The exponential rate and adaptation asymptote were found for each subject in each experiment, as well as the rate and asymptote of readaptation to normal cues. The rate of adaptation does not show any statistical dependence on experimental conditions; however, the asymptote of the best-fit slope varied with the strength of the transformation used in each experiment. This result is consistent with the hypothesis that subjects cannot adapt to a nonlinear transformation of auditory localization cues, but instead adapt to a linear approximation of the transformation. Over time, performance changes exponentially towards the best-fit linear approximation for the transformation used in a particular experiment, and the rate of this adaptation does not depend upon the transformation employed.

Adaptation to supernormal auditory localization cues as a function of rearrangement strength

The ability of humans to adapt to intermodal discrepancies is an important factor in the design of virtual environments and teleoperator systems. Previously, it was demonstrated that subjects can adapt to supernormal auditory localization cues that enhance position resolution using either standard psychophysical training methods (correct‐answer feedback) or methods traditionally employed in studies of sensorimotor rearrangement (active sensorimotor loop tasks without special cognitive feedback). In the present study, the strength of the supernormal rearrangement was parametrically varied in order to ascertain the impact on the rate and asymptote of adaptation achieved. Subjects were repeatedly tested in a hybrid real/virtual environment where auditory stimuli were synthesized (using a PC, Convolvotron and electromagnetic head tracker) for thirteen positions in the horizontal plane marked by real lights. Testing consisted of identifying the azimuth of the virtual sound source without significant head motio...

Adaptation to transformed auditory localization cues in a hybrid real/virtual environment

The ability of humans to adapt to inter‐modal discrepancies is an important factor in the design of virtual environments. In the present study, azimuthal localization cues were altered (to magnify interaural differences) relative to real proprioceptive, visual, and vestibular cues. Subjects were alternately tested and trained in hybrid real/virtual environments where auditory stimuli were synthesized (using a PC, Convolvotron, and electromagnetic head tracker) to be a 1 of 13 discrete positions marked by real lights. Testing consisted of identifying the azimuth of virtual sound sources without correct answer feedback or significant head motion. Preliminary findings on resolution and bias for a variety of different training procedures as well as a number of different transformations of the localization cues will be overviewed. [Work supported by AFOSR, Grant No. 90‐200.]

Effects of Various Types of Nonlinear Distortion upon the Intelligibility of Speech

Articulation tests were conducted under uniform conditions (same talkers, same listeners, randomized block design of experiment) to compare the effects upon intelligibility of various types and amounts of nonlinear distortion. The types included peak clipping (both symmetrical and asymmetrical), center clipping, linear rectification, and parabolic (“square-law”) rectification. The results confirm previous reports that intelligibility is highly resistant to symmetrical peak clipping (80 percent word articulation with infinite peak clipping). They show that asymmetrical peak clipping is almost entirely without detrimental effect upon intelligibility until more than half the speech wave is eliminated (98 percent word articulation with half-wave rectification). In addition, the results provide a basis for comparing effects of odd and even harmonic distortion, for comparing the effects of noise that enters the system at a point ahead of the nonlinear circuit with effects of noise that enters at a point followi...