Hearing what the patient doesn't say

For five years, research and scientific studies on vocal biomarkers are multiplying, whether that is for the diagnostic (recording used to detect the presence of an illness) or for the follow-up of patients (repeated recordings are used to measure a pathology’s evolution or the efficiency of certain treatments).

We can hear what your patient doesn’t tell you

Many start-ups have already started developing smartphone apps or websites. In the U.S., Kintsugi already offers to professionals a mental health disorders detection tool with the slogan: “We can hear what your patient doesn’t tell you.”

Ellipsis Health also targets employers on their website by saying that “early detection and a better employee engagement […] lowers the costs related to absenteeism.” Their compatriot Sonde Health developed a smartphone app about mental health wellbeing that also allows them to collect more vocal samples. They also offer biomarkers for respiratory illnesses under license.

Limited data

Vocal Biomarkers also got the attention of major pharmaceutical groups: last September, Pfizer spent $116 million to acquire ResApp Health, an Australian company with a technology able to detect 92% of COVID-19 cases with a coughing recording. And the Canadian start-up Winterlight Labs, founded in 2015, works with Genentech (Roche group) to evaluate the development of Alzheimer’s illness.

The first studies on vocal biomarkers were led 30 years ago on Parkinson’s disease, which makes a perceptible change in the voice of most patients. “Research considerably accelerated these five last years […] they benefited from smartphones generalization that allowed us to record and transmit vocal data anywhere but also from the progress of audio signal analysis and AI,” says Guy Fagherazzi.

Since then, “the pandemic has played a huge role,” says the expert. Since early 2020, a lot of studies have been launched in different countries comparing the recordings of symptomatic and asymptomatic cases to detect the illness. But there is still a lot of work to demonstrate the validity of the work and the quality of the data. Most of the existing studies were made on specific populations, with limited samples and a lot of labs failed to reproduce the same results with a larger population.

Listen up

Ethical concerns

“A lot of of data bases are too small or do not suit the studied pathologies,” explains Vincent Martin, a researcher who just devoted his doctorate thesis to vocal biomarkers for drowsiness detection. "For example, the most used corpus to detect depression in the voice was recorded with veterans suffering from post-traumatic symptoms.” A trained model on this sample will present a high error score if applied to the general population.

The use of vocal biomarkers will have to go through the the medical community and earn the patient’s trust. Its backers will have to clearly explain the benefits (high-speed, non-invasive, early diagnostic, online checkups) while guaranteeing the confidentiality and transparency of the tools.

The biggest fear is that, like facial recognition, digital tools detecting vocal biomarkers could be used without users' permission, which is concerning for technologies that are supposed to evaluate physical and mental health.