SUMMARY The organization of human auditory cortex remains unresolved, due in part to the small stimulus sets common to fMRI studies and the overlap of neural populations within voxels. To address these challenges, we measured fMRI responses to 165 natural sounds and inferred canonical response profiles (“components”) whose weighted combinations explained voxel responses throughout auditory cortex. This analysis revealed six components, each with interpretable response characteristics despite being unconstrained by prior functional hypotheses. Four components embodied selectivity for particular acoustic features (frequency, spectrotemporal modulation, pitch). Two others exhibited pronounced selectivity for music and speech, respectively, and were not explainable by standard acoustic features. Anatomically, music and speech selectivity concentrated in distinct regions of non-primary auditory cortex. However, music selectivity was weak in raw voxel responses, and its detection required a decomposition method. Voxel decomposition identifies primary dimensions of response variation across natural sounds, revealing distinct cortical pathways for music and speech.

Abstract. The sound of crashing waves, the roar of fast-moving carssound conveys important information about the objects in our surroundings. In this work, we show that ambient sounds can be used as a supervisory signal for learning visual models. To demonstrate this, we train a convolutional neural network to predict a statistical summary of the sound associated with a video frame. We show that, through this process, the network learns a representation that conveys information about objects and scenes. We evaluate this representation on several recognition tasks, finding that its performance is comparable to that of other state-of-the-art unsupervised learning methods. Finally, we show through visualizations that the network learns units that are selective to objects that are often associated with characteristic sounds.

Psychophysical experiments conducted remotely over the internet permit data collection from large numbers of participants, but sacrifice control over sound presentation, and therefore are not widely employed in hearing research. To help standardize online sound presentation, we introduce a brief psychophysical test for determining if online experiment participants are wearing headphones. Listeners judge which of three pure tones is quietest, with one of the tones presented 180° out of phase across the stereo channels. This task is intended to be easy over headphones but difficult over loudspeakers due to phase-cancellation. We validated the test in the lab by testing listeners known to be wearing headphones or listening over loudspeakers. The screening test was effective and efficient, discriminating between the two modes of listening with a small number of trials. When run online, a bimodal distribution of scores was obtained, suggesting that some participants performed the task over loudspeakers despite instructions to use headphones. The ability to detect and screen out these participants mitigates concerns over sound quality for online experiments, a first step toward opening auditory perceptual research to the possibilities afforded by crowdsourcing.

A central goal of sensory neuroscience is to construct models that can explain neural responses to natural stimuli. As a consequence, sensory models are often tested by comparing neural responses to natural stimuli with model responses to those stimuli. One challenge is that distinct model features are often correlated across natural stimuli, and thus model features can predict neural responses even if they do not in fact drive them. Here, we propose a simple alternative for testing a sensory model: we synthesize a stimulus that yields the same model response as each of a set of natural stimuli, and test whether the natural and “model-matched” stimuli elicit the same neural responses. We used this approach to test whether a common model of auditory cortex—in which spectrogram-like peripheral input is processed by linear spectrotemporal filters—can explain fMRI responses in humans to natural sounds. Prior studies have that shown that this model has good predictive power throughout auditory cortex, but this finding could reflect feature correlations in natural stimuli. We observed that fMRI responses to natural and model-matched stimuli were nearly equivalent in primary auditory cortex (PAC) but that nonprimary regions, including those selective for music or speech, showed highly divergent responses to the two sound sets. This dissociation between primary and nonprimary regions was less clear from model predictions due to the influence of feature correlations across natural stimuli. Our results provide a signature of hierarchical organization in human auditory cortex, and suggest that nonprimary regions compute higher-order stimulus properties that are not well captured by traditional models. Our methodology enables stronger tests of sensory models and could be broadly applied in other domains.