Repetition detection and rapid auditory learning for stochastic tone clouds
Stochastic sounds are useful to probe auditory memory, as they require listeners to learn unpredictable and novel patterns under controlled experimental conditions. Previous studies using white noise or random click trains have demonstrated rapid auditory learning for instances of such a class of sounds. Here, we tested stochastic sounds that enabled parametrical control of spectrotemporal complexity: tone clouds. Tone clouds were defined as broadband combinations of tone pips at randomized frequencies and onset times. Varying the density of tones covered a perceptual range from random melodies to noise. Results showed that listeners could detect repeating patterns in tone clouds at all tested densities, with sparse tone clouds being the easiest. A model estimating amplitude modulation within cochlear filters showed that repetition detection was correlated with the amount of amplitude modulation at lower rates. Rapid learning of individual tone clouds was also observed, again for all densities. Tone clouds thus provide a tool to probe auditory learning in a variety of task-difficulty settings, which could be useful for clinical or neurophysiological studies. They also show that rapid auditory learning operates over the full range of spectrotemporal complexity typical of natural sounds, essentially from melodies to noise.