An algorithm to improve speech recognition in noise for hearing-impaired listeners

Area of Science:

• Auditory Neuroscience
• Signal Processing
• Speech Communication

Background:

• Monaural (single-microphone) speech enhancement algorithms struggle to improve speech intelligibility in noise.
• Hearing-impaired (HI) listeners face significant challenges understanding speech in noisy environments.

Purpose of the Study:

• To develop and evaluate a novel binary masking algorithm for speech-noise separation.
• To assess the effectiveness of this algorithm for improving speech intelligibility, especially for HI listeners.

Main Methods:

• A binary masking algorithm was developed, estimating masks from training data separate from testing data.
• Speech was mixed with speech-shaped noise and babble at varying signal-to-noise ratios (SNRs).
• The algorithm's performance was tested using both normal-hearing and HI listeners.

Main Results:

• The algorithm successfully increased speech intelligibility across all tested conditions for both listener groups.
• Intelligibility gains were more pronounced for HI listeners, modulated backgrounds, and lower SNRs.
• Several HI listeners experienced substantial improvements, with intelligibility scores rising from near zero to over 70%.

Conclusions:

• The developed monaural binary masking algorithm effectively enhances speech intelligibility in noise.
• This algorithm offers significant benefits for HI listeners, potentially overcoming major communication barriers.
• The approach demonstrates a promising direction for practical hearing assistance devices.