Deep Learning vs. Conventional DSP: A Systematic Review on Speech Intelligibility Enhancement for Hearing Impairments

Abstract

Hearing aid devices (HAD) often fail to provide sufficient speech intelligibility in noisy environments, leading to low user adherence. While Artificial Intelligence (AI) offers adaptive signal processing, its clinical efficacy remains debated. This systematic review evaluates the impact of AI on speech intelligibility and sound quality compared to traditional digital processing. Following PRISMA guidelines, a search was conducted across specialized databases. Preliminary findings indicate that while AI architectures significantly reduce listening effort, the correlation with objective intelligibility scores varies by noise type. The certainty of evidence was assessed using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system, revealing a need for standardized clinical protocols.