Feature Extraction and Analysis of AI-Based Speech Signals for Auditory Rehabilitation
Seunghan Ha, Sangdo Lee · Asia-pacific Journal of Convergent Research Interchange · 2023
The purpose of this study was to extract and analyze features of speech signals using artificial intelligence techniques, with the goal of improving auditory rehabilitation outcomes.Wav2vec 2.0 was used to identify phonemes in 2,000 sound files recorded from individuals with hearing impairment.The average speech intelligibility score was calculated to be over 0.92 %, determined by calculating the difference between a reference sentence and the sentence obtained through speech-to-text (STT).It was possible to distinguish individual phonemes accurately from recorded sound files of speech produced by individuals with hearing impairments, and to measure the intelligibility of their speech by assessing their speech clarity.Through this experiment, we confirmed the potential to distinguish phonemes between individuals with normal hearing and those with hearing impairment, as long as the phonemes are similar to the reference sentence for individuals with normal hearing.Furthermore, it was confirmed that it is possible to differentiate the accuracy of pronunciation in speech produced by individuals with hearing impairments, which further supports the potential for assessing speech intelligibility in this population through phonetic analysis.To validate the effectiveness of artificial intelligence-based speech signal feature extraction and analysis for auditory rehabilitation, a comparative study on phoneme extraction between individuals with normal hearing and those with hearing impairment should be conducted.