Related Experiment Videos
Objective evaluation of vowel identification with the Digisonic cochlear implant
W Serniclaes1, R De Guchteneere, T Secqueville
1Laboratoire de Statistique Médicale, Ecole de Santé Publique, Université Libre de Bruxelles, Belgium.
Summary
Objective methods for evaluating cochlear implant speech recognition are crucial. The mel strategy, with more channels below 4 kHz, improved vowel classification accuracy compared to the linear strategy.
Area of Science:
- Audiology
- Speech Processing
- Biomedical Engineering
Background:
- Perceptual assessment of cochlear implant (CI) speech recognition faces challenges with patient availability and specificity.
- Objective evaluation methods using automatic speech sound classification offer a viable alternative.
Purpose of the Study:
- To evaluate different tuning strategies for the Digisonic cochlear implant using discriminant analysis.
- To assess the impact of channel frequency spacing (linear vs. mel) on vowel classification accuracy.
Main Methods:
- Utilized a dataset of 600 vocalic items from 20 French speakers, covering 10 vowel categories.
- Processed vocalic items using the Digisonic DX10 with two strategies: linear (equally spaced in Hz) and mel (equally spaced in psychoacoustic units).
- Quantified energy levels of frequency channels and applied statistical discriminant analysis for classification.
Main Results:
- The mel strategy demonstrated a higher percent correct classification (PCC) for vowels compared to the linear strategy.
- This improvement was attributed to a greater concentration of frequency channels below 4 kHz in the mel strategy.
Conclusions:
- The mel frequency spacing strategy enhances objective vowel classification performance in cochlear implants.
- Findings suggest implications for optimizing cochlear implant tuning strategies for improved speech recognition.