Artificial intelligence-assisted acoustic voice analysis as a public health tool for screening and triage of laryngeal diseases: a systematic review with qualitative synthesis

Introduction and aim. Voice disorders and laryngeal diseases impair communication, quality of life and work ability. Persistent dysphonia may also indicate glottic or laryngeal cancer. This systematic review evaluates artificial intelligence (AI)-assisted acoustic voice analysis as a non-invasive tool for screening support, triage and risk stratification of laryngeal diseases and voice disorders. Methods. The review was guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement. PubMed/MEDLINE, Scopus, Web of Science, IEEE Xplore and Google Scholar were searched using predefined Boolean strategies. The search identified 412 records; after removal of 118 duplicates, 294 titles and abstracts were screened. Eighty-two full texts were assessed, and 30 publications were included in qualitative synthesis. Meta-analysis was not performed because of heterogeneity in target conditions, voice tasks, acoustic features, artificial intelligence (AI) models and outcome metrics. Brief summary of current knowledge. AI-assisted acoustic voice analysis shows high performance in binary classification of healthy and pathological voice, especially using mel-frequency cepstral coefficients (MFCCs), perturbation measures, harmonicto-noise ratio (HNR) and spectrogram-based features. However, multiclass classification and differentiation between benign and malignant laryngeal disease remain challenging. Multimodal models using voice, demographic and clinical data appear more promising than voice-only approaches. Summary. AI-assisted acoustic voice analysis may support scalable screening, triage and telemonitoring, particularly where specialist laryngological care is limited. Implementation requires standardized recordings, external validation, explainable models, safety-netting and prospective clinical studies.