Techniques to Infer Frequency-Specific Audibility of Speech Using Single-Channel Cortical Responses

Abstract

Hearing loss is a common congenital health condition that affects audibility of speech—critical for communication development in children—in a frequency-specific manner. The use of hearing aids to amplify speech is a common intervention approach. Since hearing aids are fit within the first few months of life, there is a need to assess the efficacy of hearing aids using objective methods like electroencephalography (EEG). In this paper, six binary classification tasks are designed for frequency-specific audibility assessment using EEG-based cortical responses to speech stimuli. Three techniques, two conventional and one based on machine learning are developed for classifying the cortical responses. These techniques are compared to identify the most accurate ones under the different classification tasks. The results in this paper show that the use of machine learning offers several benefits over conventional techniques for inferring frequency-specific hearing loss using cortical responses.