Cochlear implants (CIs) are a device that can provide auditory input for individuals who are deaf, giving them access to spoken language. The vast majority of research has focused on how individuals with CIs perceive speech produced by others; however, CIs additionally provide access to auditory input from one’s own voice during speech production, known to be critical for ensuring accurate and intelligible speech. This experiment aimed to investigate whether and how typical hearing participants can learn to use CI simulated speech input for perceptual learning and sensorimotor control of speech during production. CI simulation was achieved via noise vocoding, a technique that degrades the spectral detail in a speech signal in a manner similar to a CI.

The experiment was made up of two phases. First, a perceptual learning phase tested recognition of noise vocoded sentences before and after a training task which was manipulated between groups; either a perception training task where participants listened to noise vocoded sentences while reading matching text, or a production training task where participants read aloud sentences whilst hearing their own voice noise-vocoded in real-time. In the second phase of the experiment, all participants read aloud sentences while hearing their voice noise vocoded in real-time; after a baseline phase, an additional real-time manipulation of the formants of their voice (spectral properties of speech which determine the vowel sound) was made. The direction of this formant perturbation was manipulated between groups; assignment to these groups was counterbalanced relative to assignment to training conditions in phase 1. When speaking with non-vocoded feedback, speakers typically show corrective adjustments to their production of speech sounds in order to compensate for such formant perturbations; this is known as speech motor adaptation. This process ensures that our speech productions remain on target.

Preliminary results with N = 20 (target N = 30) indicate that both the perception and production training tasks in the first phase resulted in significant improvements in recognition of noise vocoded sentences; training type however did not affect the magnitude of this improvement. For the second phase, despite trends in the right direction, speech motor adaptation in our incomplete sample was not significant at the group level in response to the formant perturbations. Once the full sample has been collected, we will also compare adaptation between the two training type groups, to test if greater experience of speaking with noise vocoded feedback results in improved adaptation.