NIPS99 workshop Neural mechanisms of music perception

Abstract of my talk:

I'll review several recent models of pitch perception, multiple pitch perception (of concurrent instruments), and source segregation. All are based on a hypothetical neural mechanism involving temporally precise inhibition. In their simplest form, the models involve a gating neuron fed by two pathways, one direct and excitatory, and the other indirect and inhibitory. All spikes arriving along the direct pathway are transmitted, unless they coincide with a spike arriving along the delayed pathway. This affects the statistics of inter-spike intervals, and suppresses the correlates of periodic sources with a period equal to the delay. Tuning this "cancellation filter" to the period of a source suppresses that source and allows other sources to emerge. The filter can also be used to estimate the period (and therefore the pitch) of a sound, by exploring the delay parameter space in search of a minimum. In that case the model is formally equivalent to the classic autocorrelation pitch perception model of Licklider. However the cancellation model can easily be extended to explain the perception of multiple pitches evoked by simultaneously playing instruments. Recent psychophysical results on the pitch shifts of mistuned harmonics can be used to predict the topology of the neural circuits involved.

(1) A. de Cheveigné (1999). "Pitch shifts of mistuned partials: a time-domain model," J. Acoust. Soc. Am. 106,887-897. [abstract]
(2) A. de Cheveigné and H. Kawahara (1999). "Multiple period estimation and pitch perception model," Speech Communication 27,175-185. [abstract and PostScript]
(3) A. de Cheveigné (1998). "The auditory system as a separation machine.", Proc. ATR workshop on events and auditory temporal structure, 1-7. [PDF]
(4) A. de Cheveigné (1998). "Cancellation model of pitch perception," J. Acoust. Soc. Am. 103,1261-1271. [abstract]
(5) A. de Cheveigné (1997). "Concurrent vowel identification III: A neural model of harmonic interference cancellation," J. Acoust. Soc. Am. 101,2857-2865. [abstract]
(6) A. de Cheveigné (1993), "Time-domain comb filtering for speech separation," ATR Human Information Processing Laboratories technical report TR-H-016.

See also my list of publications

Alain's Home page