Everyone's voice is unique. Because of our individual nuances in anatomy, it’s as distinguishable as a fingerprint. While those differences help us tell one another apart, our voices might also hold clues to detecting laryngeal cancer (cancer of the voice box).
A research team from the Department of Clinical Epidemiology at Oregon Health and Science University has shown it’s possible to detect abnormalities in the vocal folds simply by analyzing the sound of a person’s voice, in a recent study published in Frontiers in Digital Health.
While not all irregularities are dangerous, they can appear in the early stages of laryngeal cancer. Flagging them through a non-invasive AI voice test could dramatically improve treatment outcomes and survival rates.
Voice Box Cancer
The larynx, or voice box, is an intricate apparatus that sits midway down the throat. Built from cartilage, ligaments, muscles, and membranes, it transforms breath into sound. At its core are the vocal folds, which vibrate to create each person’s one-of-a-kind voice.
Though relatively rare, laryngeal cancer affected 200,000 people worldwide in 2021. Five-year survival rates vary widely (35 percent to 75 percent) depending on how early it’s caught, the cancer’s location, and how quickly treatment begins. Smoking, heavy alcohol use, and human papillomavirus (HPV) infection are among the leading risk factors.
Early detection is critical, but current methods are often invasive or require specialized equipment and expertise. That’s why researchers are exploring new, more accessible ways to spot the disease early.
Read More: Early Prostate Cancer Diagnosis Soon Possible with Simple Urine Sample
Listening for Cancer
In their recent study, the team analyzed 12,523 voice recordings from 306 people across North America. The recordings came from the public Bridge2AI-Voice dataset, which includes samples from healthy individuals as well as people with laryngeal cancer, benign vocal fold lesions, or other voice issues.
They focused on subtle acoustic markers, like pitch (mean fundamental frequency), jitter (pitch variation), shimmer (amplitude variation), and the harmonic-to-noise ratio, which compares the pure tones in a voice to background noise.
The results revealed clear differences in harmonic-to-noise ratio and pitch between men with healthy voices, men with benign lesions, and men with laryngeal cancer. No equally clear patterns emerged in women’s voices, something the researchers believe could change with a larger dataset.
The findings suggest that variations in the harmonic-to-noise ratio, in particular, could help track vocal fold lesions and detect cancer early, at least in men.
Improving AI Tools
This research was part of the Bridge2AI-Voice project within the U.S. National Institutes of Health’s “Bridge to Artificial Intelligence” initiative, a nationwide effort to apply AI to complex biomedical challenges.
After seeing success with the initial smaller sample size used in this study, the team wants to train the algorithms on even larger, professionally labeled voice datasets and test them in clinical settings, making sure the system works just as well for women as for men.
“Our results suggest that ethically sourced, large, multi-institutional datasets like Bridge2AI Voice could soon help make our voice a practical biomarker for cancer risk in clinical care,” said Phillip Jenkins, postdoctoral fellow in clinical informatics at Oregon Health & Science University, in a press statement.
Jenkins also noted that voice-based health tools are already put to the trial and he estimates similar tools for detecting vocal fold abnormalities to enter pilot testing soon.
This article is not offering medical advice and should be used for informational purposes only.
Read More: A Simple Spit Test Could Reveal Prostate Cancer, Outperforming a Common Blood Test
Article Sources
Our writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article:
