AI voice tool flags Type 2 diabetes in seconds

A new study indicates that an artificial intelligence-driven tool may assist in detecting Type 2 diabetes within seconds by analyzing subtle changes in speech. Researchers state the technology could "open a new pathway for diabetes screening," particularly for individuals who do not undergo regular health checks. However, they emphasize that the tool does not replace current blood tests used to diagnose the disease.

Speech markers linked to condition

The tool was developed by researchers at technology company Thymia and RMIT University in Melbourne, Australia. It analyzes vocal and speech variations associated with Type 2 diabetes, including increased hoarseness, vocal strain, and difficulty controlling breathing during speech. Researchers believe these changes may relate to how high blood sugar levels affect the vagus nerve, which helps control laryngeal muscles.

Additionally, people with diabetes are more prone to gastroesophageal reflux, which can irritate vocal cords and cause hoarseness. The study also notes that certain lung function disorders can impact airflow during speech.

Training data and accuracy rates

Researchers trained the tool using over 63,000 audio samples from more than 21,000 participants in the United Kingdom and the United States. They tested it using 20-second audio recordings of participants reading aloud. The model assigned higher risk scores to individuals who reported having Type 2 diabetes in 80% of cases.

In a separate analysis involving 801 people who underwent home blood tests within three months of recording their voices, the accuracy rate reached 75%. While the tool performed well across different age groups and genders, its performance declined when analyzing voices of Black participants. Researchers suggest the lower number of participants from this demographic may be one reason for the reduced effectiveness.

Screening gaps and future plans

Delayed diagnosis of Type 2 diabetes remains a significant issue, as symptoms such as fatigue and extreme thirst often appear gradually. Some individuals do not undergo routine screenings that help detect the condition. Giedrius Sibeikaitis, a researcher at Thymia, stated the technology "has the potential to transform screening and early detection processes." He explained that collecting a voice sample via phone or app could help reach people who do not regularly attend health check-ups.

Sibeikaitis stressed that the tool is not a substitute for blood testing. "Our model opens a new pathway for diabetes detection. It is not a replacement for blood tests, and should in no way discourage anyone who believes they need testing from getting it done," he said. Researchers plan to test the model in clinical settings next and evaluate its effectiveness across various community groups before broader implementation.