Researchers want to use a database of human voices to develop an AI that can recognize diseases, purely by voice. This includes serious diseases such as Alzheimer’s and cancer.
The project is funded by the US National Institutes of Health and aims to explore how far the human voice can be used as an indicator of health. Voice sound, if successful, would then rank among indicators such as body temperature and blood pressure in the diagnosis of diseases.
“The great thing about voice data is that it’s probably some of the cheapest data you can get from people,” Olivier Elemento, one of the lead researchers on the project, told tech site The Verge. ‘It is an easily available form of information that you can request from almost any patient.’
Database
It is not the first study to explore the potential of voice for diagnosis, but previous studies have been very limited in size. In the first place, work is now underway on a database, because large amounts of (labeled) voting data are not yet available, the researchers say. They want to collect this data via an app that records voice data from participants with, for example, Parkinson’s or Alzheimer’s, under the supervision of a clinician.
An AI model is then built on the basis of that dataset. What is interesting about this project is that the actual voice data remains stored in the hospital or health organization where it was collected. The computer that builds the AI model travels itself from institute to institute to avoid having to upload data. In this way, researchers want to ensure the privacy of patients.
The project is funded by the US National Institutes of Health and aims to explore how far the human voice can be used as an indicator of health. Voice sound, if successful, would then rank among indicators such as body temperature and blood pressure in the diagnosis of diseases. “The great thing about voice data is that it’s probably some of the cheapest data you can get from people,” Olivier Elemento, one of the lead researchers on the project, told tech site The Verge. ‘It is an easily available form of information that you can request from almost any patient.’ It is not the first study to explore the potential of voice for diagnosis, but previous studies have been very limited in size. In the first place, work is now underway on a database, because large amounts of (labeled) voting data are not yet available, the researchers say. They want to collect this data via an app that records voice data from participants with, for example, Parkinson’s or Alzheimer’s, under the supervision of a clinician. An AI model is then built on the basis of that dataset. What is interesting about this project is that the actual voice data remains stored in the hospital or health organization where it was collected. The computer that builds the AI model travels itself from institute to institute to avoid having to upload data. In this way, researchers want to ensure the privacy of patients.