Artificial intelligence (AI) can analyze speech patterns to accurately detect type 2 diabetes. This method can be a valuable diagnostic tool, but it comes with a warning. Medical diagnostic tools that use advanced voice analysis are becoming more precise. Analyzing […]
Artificial intelligence (AI) can analyze speech patterns to accurately detect type 2 diabetes. This method can be a valuable diagnostic tool, but it comes with a warning. Medical diagnostic tools that use advanced voice analysis are becoming more precise. Analyzing speech patterns can provide valuable insights, especially when it comes to diseases such as Parkinson’s or Alzheimer’s. Depression, post-traumatic stress disorder, and heart disease can also be detected through voice analysis. AI can even detect signs of narrowed blood vessels or exhaustion. This allows medical professionals to treat patients earlier and reduce potential risks.
According to a study published in the medical journal “Mayo Clinic Proceedings: Digital Health,” just a short voice recording is enough to accurately determine whether a person has type 2 diabetes.
This technology is designed to help identify people living with undiagnosed diabetes. Globally, around 240 million adults have diabetes and are unaware of it. Almost 90% of cases are type 2 diabetes, according to the International Diabetes Federation. People with type 2 diabetes are at increased risk of cardiovascular diseases such as heart attack, stroke, and poor circulation in the legs.
Voice analysis tests for diabetes could significantly improve the detection of this disease. Most other tests require a visit to a healthcare professional. This includes tests for fasting blood glucose (FBG), oral glucose tolerance test (OGTT), or glycated hemoglobin (A1C). The latter is performed to measure average blood sugar levels over two to three months.
How does voice analysis work?
Artificial intelligence analyzes voice changes that are not audible to the human ear. Often, recordings of phone conversations are all that the software needs for analysis. It examines factors such as speech melody, rhythm, pauses, and tone. Certain symptoms have distinctive phonetic characteristics, such as the way the vowel “A” is pronounced for five seconds. Human voice can have up to 200,000 different characteristics. AI algorithms can filter all these characteristics and identify specific vocal patterns that correspond to certain symptoms.
The newly developed artificial intelligence analyzes voice recordings lasting six to ten seconds, looking for differences in tone and voice intensity. When combined with basic health data such as age, gender, height, and weight, the program can assess whether a person has type 2 diabetes.
Its results are extremely accurate but slightly skewed depending on gender. Due to differences in voice variability between male and female speakers, the tests were accurate in 89% of cases for women and 86% for men.
Specific acoustic properties
To train the artificial intelligence, Jaycee Kaufman and her team from the Ontario Tech University in Canada recorded the voices of 267 people who either did not have diabetes or had already been diagnosed with type 2 diabetes. Over the course of two weeks, participants recorded short sentences six times a day on their smartphones. This generated over 18,000 voice samples, from which 14 acoustic properties were extracted that differed between participants with and without diabetes.
“Current detection methods can require a lot of time, travel, and costs,” said Kaufman, a research scientist at Klick Labs. “Voice technology has the potential to eliminate these barriers completely.” In the future, Klick Labs hopes to investigate whether voice analysis can also help detect other conditions such as prediabetes or hypertension.
Dangers of voice analysis
Advocates of voice analysis as a diagnostic tool highlight the speed and efficiency with which diseases can be detected using a patient’s voice. However, even though AI-supported tools can provide very specific information, a few voice samples are not enough for a definitive diagnosis. The risk of false positive results or overdiagnosis is also high. Ultimately, assessments should always be made based on human expertise.
Warning, not a medical diagnosis
This is especially true for psychological illnesses. A certain tonality of voice may be a sign of depression, for example, but only a thorough analysis by a trained human professional can provide certainty.
Abuse is not excluded
Critics and data protection advocates warn of the enormous risk of misuse of voice analysis software, for example, by employers or insurance call centers. There is a risk that voice analysis software may be used without explicit consent and that customers or employees may be harmed based on personal health information.
It would also be relatively easy to transmit sensitive medical information, hack, sell, or otherwise abuse it. However, clear regulations and limitations on voice analysis as a diagnostic tool cannot be established by scientists alone. This is the exclusive jurisdiction of policy makers.