Classification of English vowels using speech evoked potentials

Area of Science:

• Neuroscience
• Audiology
• Speech Processing

Background:

• Speech Evoked Potentials (SpEPs) are auditory brainstem responses to speech stimuli.
• Previous research indicates SpEPs contain valuable information for auditory neural processing.
• SpEPs show potential for diagnosing central auditory processing disorders and language disability, especially in children.

Purpose of the Study:

• To determine if SpEPs contain information to distinguish different speech stimuli.
• To examine spectral amplitude information of SpEPs in response to English vowels.
• To assess the utility of SpEPs for objective auditory assessment.

Main Methods:

• Investigated spectral amplitude features of Envelope Following Response (EFR) and Frequency Following Response (FFR) from SpEPs.
• Utilized SpEPs elicited by five English vowels: /a/, /e/, /ae/, /i/, /u/.
• Employed linear discriminant analysis (LDA) for classification.

Main Results:

• Achieved 78.3% classification accuracy using spectral amplitude features.
• Demonstrated that brainstem neural responses in the F0 and F1 regions of SpEPs are informative for vowel discrimination.
• Confirmed that spectral components at the fundamental frequency (F0) and first formant (F1) are key discriminators.

Conclusions:

• SpEPs provide valuable information for discriminating speech sounds at the brainstem level.
• These findings offer insights into human speech processing and auditory perception.
• SpEP analysis may lead to improved objective methods for assessing central hearing impairment.