Related Experiment Videos
A neural network model for optimizing vowel recognition by cochlear implant listeners.
C H Chang1, G T Anderson, P C Loizou
1Applied Science Department at the University of Arkansas at Little Rock, 72204, USA.
Summary
This study uses neural networks to optimize cochlear implant (CI) speech processing strategies for individual patients. The approach improved vowel recognition, particularly with a model accounting for channel interaction.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Artificial Intelligence
Background:
- Cochlear implant (CI) patient performance varies significantly.
- Optimal fitting of speech processing strategies is crucial for CI users.
- Individualized optimization is needed to enhance CI efficacy.
Purpose of the Study:
- To develop and evaluate a neural network-based approach for optimizing CI speech processing strategies.
- To personalize speech processor parameters for improved patient outcomes.
- To investigate the impact of channel interaction modeling on performance.
Main Methods:
- A two-stage neural network model was implemented.
- Stage 1: Neural network trained to mimic patient performance on vowel identification.
- Stage 2: Trained network adjusted a free parameter (weighting function) to enhance vowel recognition.
Main Results:
- Neural network models accurately replicated the performance of five Med-EI/CIS-Link implant patients.
- Optimized weighting functions improved vowel recognition performance.
- A 4% improvement in vowel performance was achieved using a weighting function that modeled channel interaction.
Conclusions:
- Neural network-based optimization is a viable method for tailoring CI speech processing strategies.
- Modeling channel interaction within weighting functions can significantly enhance CI patient performance.
- This approach holds promise for improving speech understanding in cochlear implant users.