Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: A Systematic Review

Area of Science:

• Artificial Intelligence in Medicine
• Biomedical Signal Processing
• Clinical Informatics

Background:

• Audio-based voice and speech biomarkers are emerging AI tools in medicine.
• Deep learning models offer superior performance but face clinical adoption challenges due to their black-box nature.
• Explainable AI (XAI) aims to provide transparent, understandable explanations for AI decisions.

Purpose of the Study:

• To systematically review XAI methods used for deep learning in clinical audio applications.
• To identify XAI-generated insights and assess their clinical applicability and stakeholder relevance.
• To propose recommendations for future clinical audio XAI design.

Main Methods:

• Systematic literature review following PRISMA guidelines (Jan 2015 - Feb 2025).
• Searched six major electronic databases for eligible studies.
• Thematic synthesis of results across explainability categories, clinical domains, and validation strategies.

Main Results:

• Thirty studies utilized diverse XAI methods (gradient-based, perturbation-based, attention-based, etc.).
• Applications covered voice disorders, neurodegenerative diseases, psychiatric conditions, and TBI.
• Limited quantitative validation and no human-in-the-loop evaluations with stakeholders were found.

Conclusions:

• Current XAI in clinical voice analysis lacks sufficient validation and stakeholder focus.
• There is a need for validated, audio-aware, and stakeholder-centered XAI approaches.
• Future XAI development must prioritize clinical applicability and trustworthiness for adoption.