From Microphone to Phoneme: An End-to-End Computational Neural Model for Predicting Speech Perception With Cochlear...

Area of Science:

• Computational neuroscience
• Bioengineering
• Auditory neurophysiology

Background:

• Neuroprostheses, like cochlear implants (CI), are vital for restoring function.
• Computational models accelerate the development and optimization of neuroprostheses.
• Predicting speech perception in CI users is crucial for improving device efficacy.

Purpose of the Study:

• To develop an end-to-end computational model for predicting speech perception in cochlear implant users.
• To utilize artificial neural networks and biological system models for virtual prototyping.
• To analyze information degradation within the CI processing chain.

Main Methods:

• Integrated CI signal processing, a finite element model of the electrically-stimulated cochlea, and an auditory nerve model.
• Employed an automatic speech recognition neural network to extract phoneme-level speech perception from neural responses.
• Validated model predictions against human CI listener data.

Main Results:

• The model accurately predicted speech perception and misperception patterns observed in human CI users.
• It captured phoneme-specific perceptibility differences and the impact of stimulation parameters and noise.
• Information transmission analysis revealed the electrode-neural interface as the primary information bottleneck.

Conclusions:

• The developed end-to-end model successfully replicated phoneme-level CI speech perception.
• It provided a quantitative measure of information degradation throughout the CI processing chain.
• This modeling approach holds significant promise for advancing neuroprosthesis development and optimization.