Perception of Coloration in Filtered Gaussian Noise—Short-Time Spectral Analysis by the Ear

Abstract

Frequency-response irregularities of electroacoustic components (loudspeakers, microphones, etc.) are usually well audible when the response irregularity exceeds several decibels. In contrast, the considerable frequency irregularity of rooms (10 db or more) is generally not perceived when listening in rooms to speech or noise. To resolve this paradox, the following hypothesis has been made: The ear is not directly sensitive to the frequency irregularity of a “filter” inserted between it and the sound source, but rather to a smoothed version of its frequency response, corresponding to the ear's inability to resolve fine detail of irregular frequency responses. (The limited frequency resolution is a necessary consequence of the short-time spectral analysis performed by the ear.)—Experiments with filtered Gaussian noise have confirmed this hypothesis for a variety of periodic frequency responses with different frequency irregularities and peak spacings. Investigations with aperiodic (more roomlike) frequency responses are in progress. Test tapes for the subjective tests were prepared by a digital computer. (The filtering by digital computer is particularly convenient because of the difficulty of instrumenting filters with the precision required in these experiments.)—The most interesting result of the subjective tests is the time-weighting function (“time window”) employed by the ear model in its short-time spectral analysis. Whether the time-weighting function obtained in noise-coloration experiments is applicable to other experiments involving temporal summations is not yet known.