Abstract:
Most studies of auditory recognition and identification have employed either speech stimuli or nonspeech sounds generated in the laboratory (e.g., tones of various frequencies, tonal patterns, click trains). The present study employed 25 naturally occurring complex sounds (obtained from a commercial sound-effects library), such as those produced by doors closing, babies crying, helicopters in flight, and other familiar events. These sounds, equated for peak levels, were recorded with a background of broadband noise. The recorded sounds were presented to groups of six to eight listeners in both open- and closed-set formats (with the list of responses displayed continuously in the latter). Confusion matrices were generated using a wide range of event-to-noise (Ev/N) ratios. Two frequent confusions were identified for each item, and were used to create a three-alternative forced-choice test. Eight values of Ev/N were selected for each item in an effort to achieve uniform item identifiability. Average Ev/N required to achieve 50-percent-correct recognition, the slopes of psychometric functions, and the distribution of these measures among normal-hearing listeners are compared to the corresponding measures for speech identification. [Work supported by AFOSR and NIDCD.]