Automated Speech Intelligibility Assessment Using AI-Based Transcription in Children with Cochlear Implants, Hearing...

Area of Science:

• Audiology
• Speech-Language Pathology
• Artificial Intelligence in Healthcare

Background:

• Speech intelligibility (SI) is crucial for children's communication and social development, particularly those with hearing loss.
• Current SI assessment methods face logistical and methodological challenges, limiting their clinical use.
• Children with hearing loss, including those using cochlear implants (CI) or hearing aids (HA), require accessible SI evaluation tools.

Purpose of the Study:

• To evaluate the accuracy and consistency of an AI-based transcription model for assessing SI in young children.
• To compare the AI model's performance against naïve human listeners.
• To analyze AI-driven SI assessment across children with CI, HA, and normal hearing (NH).

Main Methods:

• 580 speech samples from 58 five-year-old children (CI, HA, NH groups) were transcribed by three naïve listeners and an AI model.
• Word-level transcription accuracy was assessed using Bland-Altman plots, intraclass correlation coefficients (ICCs), and word error rate (WER).
• Statistical analyses compared AI performance with human listeners across different hearing groups.

Main Results:

• The AI model showed high consistency with human listeners, with minimal bias ( < 6% outside 95% limits of agreement).
• Intraclass correlation coefficients (ICCs) exceeded 0.9 for all groups, indicating strong agreement.
• Word error rate (WER) confirmed AI-human alignment, with children with CIs demonstrating better SI than those with HAs.

Conclusions:

• AI-based transcription provides a reliable and objective method for SI assessment in young children.
• The AI model's performance aligns with human listeners, supporting its clinical and home-based application.
• This technology can aid early intervention and ongoing speech development monitoring for children with hearing impairments.