When blood sugar levels go high, so does voice pitch, according to a new study. It paves the way for detecting type 2 diabetes or monitoring diabetics’ blood sugars simply by talking into a smartphone and having AI analyze their speech.
For type 2 diabetics, insulin resistance and a pancreas that gradually loses the capacity to produce enough insulin results in deranged blood sugar levels. The earlier the condition is diagnosed, the better for avoiding long-term complications. For a long time, science has been working on accurately and effectively diagnosing type 2 diabetes, from checking blood levels of a specific protein to detecting the condition via a smartphone camera.
Now, a team of digital healthcare innovators at Klick Labs has identified a new, non-invasive method of diagnosing and monitoring diabetes: voice pitch.
“By establishing a significant positive association between glucose levels and fundamental frequency [of the voice], our study provides compelling justification for more research on using voice to predict and monitor glucose levels,” said Jaycee Kaufman, lead scientist at Klick Labs and the study’s lead and corresponding author. “Whereas current glucose monitoring methods are often invasive and inconvenient, voice-based glucose monitoring could be as easy as talking into a smartphone, which would change the game for the estimated 463 million people around the world living with type 2 diabetes.”
The fundamental frequency, or F0, of the complex speech tone is just another term for pitch. It's been hypothesized that glucose levels affect voice characteristics primarily stemming from Hooke’s law, as applied to the vocal cords, which states that variations in the tension, mass, or length of the cords, influenced by glucose levels in the body, alter their vibrational frequency. So, the researchers set out to test this hypothesis.
They recruited 505 participants and divided them into three groups: 242 non-diabetics, 89 prediabetics, and 174 type 2 diabetics. All participants were fitted with a continuous glucose monitor (CGM) and instructed to record their voices in a quiet environment using a custom smartphone app up to six times daily for two weeks. The participants were asked to recite a fixed phrase – “Hello, how are you? What is my glucose level right now?” – to capture the natural frequency of their voices during speech.
At the conclusion of the study, all voice recordings were paired with participants’ closest-in-time CGM data. As glucose levels were measured every 15 minutes, this allowed all the voice recordings to be within 7.5 minutes of a glucose recording.
The researchers found a significant association between voice pitch and blood glucose levels. Analysis revealed a linear relationship; an increase in one corresponded with an increase in the other.
“Overall, the frequency of the voice has a small but significant relationship to glucose levels when evaluated within an individual,” the researchers concluded. However, they noted that pitch alone was unlikely to be able to predict blood glucose levels despite the linear relationship between the two. “Other vocal features are likely necessary to build a successful prediction model,” they said.
The researchers also note an important caveat to their findings: vocal parameters, specifically voice pitch, can be affected by external factors such as emotional and psychological states, respiratory infections, and allergies.
The present study is another step in Klick Labs’ research into the detection and management of diabetes using voice tech and machine learning. In 2023, it found distinct voice differences between type 2 diabetics and those without the condition and that AI, using this and other predictors, had potential as a pre-screening or monitoring tool.
The study was published in the journal Scientific Reports.
Source: Klick Health
