Identification of brief auditory patterns

Abstract

Listen

Translate article

The identification of brief nonspeech sounds poses a challenging problem for machine-based recognition systems, especially at low signal-to-noise ratios. In contrast, human observers are thought to be remarkably adept at such tasks. However, there are few systematic studies in the literature that have attempted to quantify human psychophysical performance, or the factors that influence performance, for the identification of brief nonspeech sounds in noise. This study examined human identification of brief sequences of tones arranged in six frequency patterns in masked conditions. The stimuli and methods have been described previously [Kidd, Jr. et al., J. Acoust. Soc. Am. 98, 1977–1986 (1995)]. Several acoustical factors were varied including pattern duration, signal-to-noise ratio, and range of frequencies forming the patterns. Identification performance declined as duration was shortened below 100 ms even when signal energy, or detectability, were equated. For brief patterns at low signal-to-noise ratios, increasing the range of frequencies comprising the patterns improved identification performance. These results provide insight into the limits of human performance in the identification of brief sounds and may prove useful in the design of algorithms for machine-based recognition. [Supported by ONR and NIH.]