Objective discrimination of bimodal speech using frequency following responses

Area of Science:

• Auditory Neuroscience
• Speech Processing
• Machine Learning in Audiology

Background:

• Bimodal hearing, combining cochlear implants (CI) with contralateral hearing aids, offers superior speech recognition compared to CI alone.
• Predicting individual success in bimodal hearing remains challenging, with potential drivers including fundamental frequency (f0) and fine structure cue extraction.
• Frequency following responses (FFR) may reflect neural encoding of these critical speech cues in bimodal listening conditions.

Purpose of the Study:

• To parametrically investigate the neural encoding of f0 and F1 in simulated bimodal speech.
• To objectively discriminate FFRs in bimodal conditions using machine learning algorithms.
• To determine if FFRs can predict perceptual benefits in bimodal hearing.

Main Methods:

• Simulated bimodal hearing conditions were created using a vocoder and low-pass filters across five acoustic bandwidths.
• Frequency following responses (FFRs) were evoked using three vowels (/ε/, /i/, /ʊ/) with identical f0.
• Machine learning was employed to classify FFRs, and perceptual performance was assessed using the BKB-SIN test.

Main Results:

• Neural encoding of f0 and F1 components in FFRs improved with increased acoustic bandwidth in the simulated non-implanted ear.
• Increased spectral differences in FFRs, correlating with bandwidth, led to more accurate classification and discrimination via machine learning.
• Enhanced neural encoding of f0 and F1 predicted perceptual bimodal benefit in speech-in-noise tasks.

Conclusions:

• Frequency following responses (FFR) show enhanced neural representation of speech cues with increasing acoustic bandwidth in simulated bimodal hearing.
• Machine learning can effectively discriminate FFRs, reflecting spectral differences crucial for bimodal speech perception.
• FFR holds promise as an objective tool for assessing individual variability and predicting outcomes in bimodal hearing.