Informational masking by everyday sounds

Tone-in-noise detection is severely degraded when only a few spectral components of the noise are presented at random on each trial [Neff and Green, Percept. Psychophys. 41, 409-415 (1987)]. The elevations in threshold are attributed to uncertainty regarding the noise caused by sparse sampling of noise components-informational masking is the term used to describe the result. The present experiment was undertaken to determine how informational masking is affected when sparse sampling is from a set of common everyday sounds instead of noise. On each presentation a different masker waveform of constant total power was synthesized from the magnitude and phase of a fixed number, m, of spectral components (m = 2-921 across blocks of trials). The components were selected at random from 1 of 50 common environmental sounds (e.g., baby crying, door slamming, phone ringing), or 1 of 50 samples of Gaussian noise. Masked thresholds for a 1.0-kHz signal in the presence of both types of maskers were obtained using an adaptive, two-interval, forced-choice procedure. Results with noise replicated earlier, results showing largest elevations in threshold for 10-20 sampled components. Results with everyday sounds showed a similar pattern with thresholds elevated above those for noise by as much as 10 dB at the larger values of m. The differences in masked thresholds were systematically related to differences in the ensemble variance of masker spectra, as predicted by a model previously applied to noise [Lutfi, J. Acoust. Soc. Am. 94, 748-758 (1993)]. Not predicted was the result of a subsequent trial-by-trial analysis in which 9-11 dB less masking was observed for samples from everyday sounds rated as easily recognized by listeners. The results suggest that listeners fail to exploit lawful dependencies among spectral components of everyday sounds to aid detection, unless enough information is available for the sound to be easily recognized.