Title : AI-derived voice biomarkers for non-invasive diagnosis of voice and neurological disorders
Abstract:
Background: Advances in artificial intelligence (AI) have enabled the identification of subtle acoustic signatures in human voice. The concept of “voice biomarkers” refers to quantifiable vocal characteristics that can distinguish between normal and pathological conditions. This study explores the role of AI-driven voice analysis in diagnosing a spectrum of laryngeal and neurological disorders using non-invasive methods.
Objectives: To identify and validate AI-based voice biomarkers across various voice disorders and neurological conditions, and to evaluate their potential as a primary diagnostic and predictive screening tool.
Methods: A total of 760 subjects were analyzed across multiple categories: normal voice (n=100), puberphonia (n=100), hyperfunctional voice disorders (n=100), hypofunctional voice disorders (n=100), thyroid-related voice changes (n=100), laryngitis (n=100), vocal cord paralysis (n=10), vocal cord polyp (n=10), vocal nodule (n=10), papilloma (n=10), Parkinson’s disease (n=10), and dementia (n=10). Voice samples were recorded and analyzed using a multi-frequency voice pitch analyzer. Acoustic parameters including pitch range, jitter, shimmer, and harmonics-to-noise ratio (HNR) were extracted. AI algorithms were applied to identify distinct biomarker patterns associated with each condition. These findings were subsequently correlated with clinical and endoscopic evaluations to validate diagnostic accuracy.
Results: Distinct voice biomarker patterns were identified for each category. Normal voices demonstrated stable baseline acoustic signatures, while puberphonia exhibited characteristic high-pitch deviations. Hyperfunctional disorders showed elevated pitch and strain patterns, whereas hypofunctional conditions demonstrated reduced pitch and diminished acoustic energy. Thyroid disorders and laryngitis presented with altered and irregular frequency patterns. Structural lesions such as vocal cord paralysis, polyps, nodules, and papilloma revealed significant deviations and condition-specific variations in biomarkers. Neurological conditions, including Parkinson’s disease and dementia, showed unique alterations reflecting motor and cognitive influences on phonation. Overall, AI-based classification demonstrated strong potential in differentiating between these conditions based solely on voice characteristics.
Conclusion: AI-derived voice biomarkers provide a promising, non-invasive, and scalable approach for diagnosing a wide range of voice and neurological disorders. This method shows potential not only for early detection but also for predictive screening before clinical manifestation. 2 Integration of AI with acoustic analysis could significantly enhance diagnostic accuracy in otolaryngology and neurology.
Future Directions: Further research should focus on expanding datasets, improving AI model robustness, conducting large-scale clinical validation, and integrating voice analysis with other diagnostic modalities. The development of predictive models may enable early identification of preclinical and potentially pre-cancerous conditions, facilitating timely intervention.
Keywords: Voice biomarkers, Artificial Intelligence, Voice disorders, Puberphonia, Acoustic analysis, Early diagnosis, Predictive healthcare
