Suppressing reverberation in cochlear implant stimulus patterns using time-frequency masks based on phoneme groups

Area of Science:

• Auditory Neuroscience
• Signal Processing
• Speech Communication

Background:

• Cochlear implant (CI) users struggle with speech perception in reverberant conditions.
• Time-frequency masking is a common strategy to suppress reverberation but is challenging due to signal variability.
• Previous phoneme-based algorithms showed promise assuming ideal phoneme knowledge.

Purpose of the Study:

• To investigate a real-time feasible phoneme-based mask estimation algorithm for CI users.
• To evaluate the effectiveness of using phoneme classifier predictions to select mask estimation models.
• To assess the benefits of causal feature extraction for real-time implementation.

Main Methods:

• Developed a phoneme-based mask estimation algorithm using phoneme classifier predictions.
• Employed causal features within the CI processing framework for both classifier and mask estimation.
• Conducted experiments with normal-hearing listeners using an acoustic model of CI processing.

Main Results:

• The phoneme-specific algorithm, using predicted phonemes, provided benefits in reverberant environments.
• The majority of normal-hearing listeners demonstrated improved speech understanding with the phoneme-specific approach.
• Real-time feasibility was maintained through causal feature extraction.

Conclusions:

• A real-time phoneme-based mask estimation strategy can significantly improve speech intelligibility for CI users in reverberant settings.
• The use of phoneme classification to guide mask selection is a viable approach for enhancing CI performance.
• Causal processing ensures the algorithm's applicability in real-time CI devices.