Real-time contrast enhancement to improve speech recognition

Area of Science:

• Speech processing
• Auditory perception
• Signal processing

Background:

• Speech perception in noise is challenging for individuals with hearing loss.
• Existing hearing aid technologies struggle to fully restore speech clarity, especially for consonants.
• Spectral smearing and talker variability further degrade speech intelligibility.

Purpose of the Study:

• To evaluate the efficacy of a real-time Contrast Enhancement (CE) algorithm.
• To determine if CE processing improves speech identification in quiet and noise.
• To assess the impact of CE on phonetic feature perception, particularly place of articulation.

Main Methods:

• Developed and implemented a Contrast Enhancement (CE) algorithm using dynamic compressive gain and lateral inhibitory sidebands.
• Utilized a modified winner-take-all circuit to sharpen the dynamic spectrum.
• Tested normal-hearing listeners on identifying spectrally smeared consonants (VCVs) and vowels (hVds) in quiet and noise conditions.

Main Results:

• The CE algorithm significantly improved consonant and vowel identification, particularly in noisy environments.
• The degree of improvement was independent of spectral smearing extent or talker characteristics.
• Place of articulation was the phonetic feature showing the most consistent improvement for consonants.

Conclusions:

• The Contrast Enhancement (CE) algorithm effectively sharpens speech dynamics, enhancing intelligibility.
• CE processing offers a promising approach for improving speech perception in hearing aid users.
• The algorithm's effectiveness in improving place-of-articulation perception addresses a key challenge for sensorineural hearing loss listeners.