Description

With the objective of using vocal biomarkers for diagnosis, risk prediction/stratification and remote monitoring of various clinical outcomes and symptoms, there is a major need to develop surveys where audio data and clinical, epidemiological and patient-reported outcomes data are collected simultaneously.

The objectives of CoLive Voice are:

• To launch an international anonymized survey where vocal recordings are associated with large validated clinical and epidemiological data, in the context of various chronic diseases or frequent health symptoms in the general population
• To extract audio features and train supervised machine learning models to identify key candidate vocal biomarkers of the aforementioned chronic conditions or related symptoms.

Participants will be recruited online and will complete the survey using a web application.

They will first answer a detailed questionnaire on their health status and then do 5 different voice records:

1. read a 30 sec prespecified text (from the Human Rights Declaration),
2. sustain voicing the vowel /aaaaaa/ as long and as steady as they can at a comfortable loudness
3. cough 3 times
4. breath in and out deeply 3 times
5. Count from 1 to 20 at a normal speed

Vocal records will be pre-processed and converted into features, meaning the most dominating and discriminating characteristics of a vocal signal. Following the selection of features, machine or deep learning algorithms will be trained to automatically predict or classify the clinical, medical or epidemiological outcomes of interest, from vocal features alone or in combination with other health-related data.