AI Analysis of Voice Recordings Could Help Detect Laryngeal Cancer

Researchers have shown that acoustic features of speech, particularly the harmonic-to-noise ratio and its variation, could help to detect abnormalities in people's vocal folds.

The study, published in Frontiers in Digital Health, analyzed voice recordings and was able to distinguish clear differences between the voices of healthy men, those with vocal fold legions and those with laryngeal cancer. These findings could open the door to a potential AI-powered tool for early vocal fold irregularity detection.

Diagnosing voice box conditions

Laryngeal cancer, affecting the voice box, is a significant health concern. In 2021, it was estimated to affect 1.1 million people globally, with around 100,000 deaths. Major risk factors include smoking, heavy alcohol consumption, and infection with human papillomavirus. Survival rates vary widely, from 35% to 78% over five years, depending on the stage and tumor location.

Diagnosis currently relies on video nasal endoscopy and biopsies, which are invasive and require specialist access, potentially delaying treatment. Detecting lesions from voice recordings could provide a non-invasive, faster screening option for at-risk patients.

The research team, part of the "Bridge2AI-Voice" project within the US National Institutes of Health’s Bridge to Artificial Intelligence consortium, analyzed 12,523 voice recordings from 306 participants in the Bridge2AI-Voice dataset. A subset of recordings came from patients with laryngeal cancer, benign vocal fold lesions, spasmodic dysphonia, or unilateral vocal fold paralysis.

Acoustic features examined included:

• Pitch – the mean fundamental frequency 
• Jitter – variation in pitch during speech
• Shimmer – variation in loudness
• Harmonic-to-noise ratio – a measure of the relation between harmonic and noise components of speech

Significant differences in harmonic-to-noise ratio and pitch were identified among men with healthy voices, benign lesions and cancer. No similar differences were detected in women, though researchers believe that a larger dataset of female voices lead to different results.

Potential clinical applications

The findings suggest that tracking changes in harmonic-to-noise ratio could help monitor lesion progression and identify early-stage laryngeal cancer in men. Though the authors emphasised that larger, more diverse datasets and clinical validation are still needed before the approach can be widely implemented.

“Here we show that with this dataset we could use vocal biomarkers to distinguish voices from patients with vocal fold lesions from those without such lesions,” said Dr Phillip Jenkins, a postdoctoral fellow in clinical informatics at Oregon Health & Science University, and the study’s corresponding author.

“To move from this study to an AI tool that recognizes vocal fold lesions, we would train models using an even larger dataset of voice recordings, labeled by professionals. We then need to test the system to make sure it works equally well for women and men,” added Jenkins.

If successful, such AI tools could be integrated into triage systems, helping to triage patients at risk for laryngeal cancer based on their voice.

“Voice-based health tools are already being piloted. Building on our findings, I estimate that with larger datasets and clinical validation, similar tools to detect vocal fold lesions might enter pilot testing in the next couple of years," predicted Jenkins.

Reference: Jenkins P, Harrison R, Bedrick S, et al. Voice as a biomarker: exploratory analysis for benign and malignant vocal fold lesions. Front Digit Health. 2025;7. doi: 10.3389/fdgth.2025.1609811

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.

This content includes text that has been generated with the assistance of AI. Technology Networks' AI policy can be found here.