Signal detection as a function of relative acoustic entropy.
Christian E. Stilp, Michael Kiefte, Keith R. Kluender · The Journal of the Acoustical Society of America · 2010
Natural sounds are far from random. Instead, they possess structure that reflects physical constraints on sound-producing objects and events. Listeners likely exploit such structure (predictability and redundancy) when perceiving complex natural sounds. The present experiments investigate listeners’ ability to discover and use structure in novel sounds to identify a repeating target sound against a competing background of sounds. Complementary signal-processing strategies independently varied acoustic entropy (complement to redundancy) in frequency or time. Novel sounds were generated by stretching the entropy in a restricted frequency band (instantaneous frequency dilation) or temporal slice (phase vocoding) of pink (1/f) noise across the bandwidth or duration, respectively, of an unstretched noise sample. Listeners matched a probe sound to a target sound (1%, 3.2%, or 10% entropy of pink noise) that repeats amidst distractor sounds (1%, 10%, or 100% entropy) at 0-dB SNR. In separate experiments varying entropy in frequency or time, identification of target sounds depended critically on the difference in entropy across targets and distractors. Listeners were more proficient identifying lower-entropy targets against higher-entropy distractors but not lower-entropy distractors. Similarly, higher-entropy targets were better identified amidst lower-entropy distractors. Potential implications for cochlear implant processing will be discussed. [Work supported by NIDCD.]