Individual differences in human frequency-following response predict pitch labeling ability

Abstract

The frequency-following response (FFR) provides a measure of phase-locked auditory encoding in humans and has been used to study subcortical processing in the auditory system. While effects of experience on the FFR have been reported, few studies have examined whether individual differences in early sensory encoding have measurable effects on human performance. Absolute pitch (AP), the rare ability to label musical notes without reference notes, provides an excellent model system for testing how early neural encoding supports specialized auditory skills. Results show that the FFR predicts pitch labelling performance better than traditional measures related to AP (age of music onset, tonal language experience, pitch adjustment and just-noticeable-difference scores). Moreover, the stimulus type used to elicit the FFR (tones or speech) impacts predictive performance in a manner that is consistent with prior research. Additionally, the FFR predicts labelling performance for piano tones better than unfamiliar sine tones. Taken together, the FFR reliably distinguishes individuals based on their explicit pitch labeling abilities, which highlights the complex dynamics between sensory processing and cognition.

Figure 1

Spread of Behavioral Data. Individual data points are provided for individual subjects. Red circles represent individuals who self-report as an AP possessor, while turquoise triangles represent other musicians. (a) Comparison of performance on the AP sine tone conservative measure compared to performance on the AP piano tone conservative measure. (b) Performance on the pitch adjustment task. (c) Performance on the just-noticeable-difference task.

Figure 2

Power Spectra of Stimuli and of Frequency-Following Responses. The nearest integer frequency to the harmonics of the stimulus is marked on each plot, except for the speech stimulus, in which every other harmonic is marked to avoid visual clutter. The EEG spectra are corrected for 1/f frequency drop-off here for visualization, but uncorrected values were used for analysis.

Figure 3

Performance of Lasso Regression Models Using FFR to Different Stimuli as Predictors. (a, b) For each model, a correlation between the model’s predictions and true AP sine and piano performance was computed on a test set (data points not seen by the model during training) for each of 1000 cross-validation runs as an estimate of how well the model generalizes. See Eqs. (1–3) for final model specifications. (c) Predicted AP sine and piano performance values based on complex tone FFR plotted against actual, observed AP performance. Red dots represent subjects who self-reported as AP possessors. (d) Predicted AP performance values based on piano tone FFR plotted against observed AP performance. Red dots represent subjects who self-reported as AP possessors. (e) Predicted AP performance values based on speech /da/ FFR plotted against observed AP performance. Red dots represent subjects who self-reported as AP possessors.

Figure 4

Predictive Performance of Piano FFR on Sine Tones and Piano Tones separately. (a, b) Correlation between the predicted pitch labelling performances and the true pitch labelling performances on a test set are shown for 1000 cross-validation runs. The FFR to the piano tone predicts pitch labelling performance for piano tones better than it does for sine tones. (c) Predicted pitch labelling performance on the piano tones plotted against actual, observed pitch labelling performance. Red dots represent subjects who self-reported as AP possessors. (d) Predicted pitch labelling performance on the sine tones plotted against observed pitch labelling performance. Red dots represent subjects who self-reported as AP possessors.

Figure 5

Predictive Performance of Behavioral Tests, Piano FFR, and a Combined Model on Pitch Labelling Performance. (a, b) Correlation between the predicted pitch labelling performances and the true pitch labelling performances on a test set are shown for 1000 cross-validation runs. The FFR to the piano tone predicts pitch labelling performance better than the behavioral tests as well as the combined model. (c) Predicted pitch labelling performance based on the behavioral tests plotted against actual, observed pitch labelling performance. Red dots represent subjects who self-reported as AP possessors. (d) Predicted pitch labelling performance based on a combined model of both behavioral tests and the piano FFR plotted against observed pitch labelling performance. Red dots represent subjects who self-reported as AP possessors. (e) Predicted pitch labelling performance based on the piano FFR plotted against observed pitch labelling performance. Red dots represent subjects who self-reported as AP possessors.