Cortical encoding of pitch: recent results and open questions

Abstract

It is widely appreciated that the key predictor of the pitch of a sound is its periodicity. Neural structures which support pitch perception must therefore be able to reflect the repetition rate of a sound, but this alone is not sufficient. Since pitch is a psychoacoustic property, a putative cortical code for pitch must also be able to account for the relationship between the amount to which a sound is periodic (i.e. its temporal regularity) and the perceived pitch salience, as well as limits in our ability to detect pitch changes or to discriminate rising from falling pitch. Pitch codes must also be robust in the presence of nuisance variables such as loudness or timbre. Here, we review a large body of work on the cortical basis of pitch perception, which illustrates that the distribution of cortical processes that give rise to pitch perception is likely to depend on both the acoustical features and functional relevance of a sound. While previous studies have greatly advanced our understanding, we highlight several open questions regarding the neural basis of pitch perception. These questions can begin to be addressed through a cooperation of investigative efforts across species and experimental techniques, and, critically, by examining the responses of single neurons in behaving animals.

Fig. 1

Some examples of periodic sounds. Each row shows the spectrum (left panel) and waveform (right panel) of a sound with an F0 of 400 Hz: (a) Pure tone; (b) Harmonic tone complex, containing tones at 400, 800 and 1200 Hz; (c) sinusoidally amplitude modulated (SAM) tone, with a carrier of 1000 Hz modulated by 400 Hz; (d) A train of clicks, with one click presented once every 2.5 ms; and (e) iterated rippled noise (IRN), with a delay period of 2.5 ms and 20 iterations.

Fig. 2

Ferrets’ discrimination performance on two F0 discrimination tasks. (a) Pitch direction judgment performance of one ferret on a two-alternative forced choice task (black dots). On each trial, a reference artificial vowel (F0 = 400 Hz, black open circle) was presented, followed by a second, target vowel of different F0 (x-axis). The ferret indicated, by water spout choice (y-axis), whether the target vowel was higher or lower in pitch than the reference. The black curve is a probit fit to the ferrets’ spout choices. (b) The mean (+standard deviation) of ferrets’ Weber fractions on two F0 discrimination tasks are shown. The Weber fractions of 4 ferrets were measured on the pitch direction judgment task described in (a), using references between 350 and 450 Hz (left; n = 6 thresholds). The Weber fractions of 3 further ferrets were measured on a go/no-go pitch change detection task (right). On each trial, the ferret was required to release a water spout when the F0 of a sequence of 400-Hz vowels changed. Detection of pitch increases and decreases were tested in separate sessions (n = 6 thresholds in total).

Fig. 3

Neurometric analysis of how well monotonic auditory cortical codes of artificial vowel F0 support pitch discrimination judgments (modified from Bizley et al., 2010). (a) Neurometric “F0 discrimination” curves for a population of 26 auditory cortical neurons that were simultaneously recorded in an anaesthetized ferret. The light gray curves show neurometrics for each individual neuron, while the black curve shows the neurometric performance based on the response of the population of 26 cells. The white circle indicates the reference value. (b) The scatter plot compares the slopes of neurometric curves for populations of auditory cortical neurons when calculated using either the number of spikes (x-axis) or relative first-spike latency (y-axis) as a response. (c) A comparison of ferrets’ psychometric and auditory cortical neurometric sensitivity. The black line shows the mean psychometric slopes of ferrets on a two-alternative forced choice pitch direction judgment task, across a range of reference F0s. The symbols plot the neurometric slopes for populations of auditory cortical neurons. Different symbols are used for populations from different cortical areas, as shown on the legend.

Fig. 4

Proportions of cortical neurons modulated by the pitch, timbre or azimuth of complex sounds. In each panel, the proportion of ferret auditory cortical neurons with spike rates that are significantly modulated by the pitch (solid line), timbre (dashed line), or azimuth location (dotted line) of artificial vowels is indicated. Mutual Information was calculated for spike counts within 20 ms time bins, across the duration of the response. The significance of mutual information was determined using the 95% confidence interval of bootstrapped, “scrambled” responses (as described in Panzeri et al., 2007). This is compared in 20 ms time bins, across the duration of the response. The five panels, from top to bottom, show sensitivity across five cortical fields: A1 (primary auditory cortex), AAF (the Anterior Auditory Field), PPF (the Posterior Pseudosylvian Field), PSF (the Posterior Suprasylvian Field), and ADF (the Anterior Dorsal Field).