Auditory cortical predictions of vocal feedback are task-dependent

Caroline Niziolek, UCSF Dept of Otolaryngology

 

How does the brain evaluate the success of a motor act? An action’s sensory consequences, such as the auditory feedback heard while speaking, are thought to be predicted by the motor system. If the feedback matches the prediction, the neural response in sensory cortices is suppressed, a phenomenon known as speaking-induced suppression (SIS). However, it is unclear how this suppression occurs neurally, and what sensory parameters are compared to evaluate a “match.” For example, when speaking English, the pitch of the voice has much more freedom to vary than when singing; is pitch encoded in the auditory prediction in the same way during these two tasks? Here, we used natural speech production to probe the nature of internal predictions and how they are encoded in auditory cortex.

We used magnetoencephalography (MEG) to measure how SIS varied over repeated word productions in two different contexts.  In Experiment 1, ten subjects produced 200 repetitions of three different words; in Experiment 2, ten subjects were cued by tonal prompts to produce 150 repetitions of a single word on three different pitches. These productions were then played back to the subjects. SIS was defined as the suppression of the auditory M100 response to spoken or sung words relative to the playback condition.

When the prediction does not encode pitch, SIS should not change across pitch space: every production will be equally accurate. However, when the prediction does encode a pitch target, then SIS should be strongest at the center of the pitch distribution (most accurate pitches) and attenuated at the periphery (sharpest and flattest attempts at producing a pitch), where the feedback least matches the prediction. We found that pitch encoding depended on task context: SIS varied across utterances only in Experiment 2, in which pitch was an explicit target. This is consistent with a forward model in which the auditory prediction is weighted by acoustic parameters that contribute to a higher-level goal.