Smarty4Covid Dataset

Overview

The idea of harnessing the power of Artificial Intelligence (AI) and m-health towards detecting new bio-markers indicative of the onset and the progress of respiratory abnormalities/conditions, has motivated several research endeavours in applying signal analysis and AI on audio recordings of cough, voice and breath with the ultimate goal to detect innovative COVID-19 related bio-markers. The Smarty4Covid dataset contains audio signals of cough, deep breathing and regular breathing as recorded by means of different devices through following a crowd-sourcing approach supported by a responsive web-based application. In addition, it includes self-reported information, such as demographics, symptoms and results of COVID-19 tests.
 

Data structure

The main directory is organized as shown below. It contains sub-directories, each including the submissions of a specific participant. Each participant directory is named after the unique participant id and it contains a unique json file (“demographics_underlying_conditions.json”) containing information about demographics (e.g. BMI, age group, sex, presence of potential underlying medical conditions). Each submission is contained in a separate sub-directory, inside the participants’ directory, named after the unique submission id and it includes:

1. audio recordings of cough (“audio.cough.mp3”), deep breathing (“audio.breath_deep.mp3”) and regular breathing (“audio.breath_regular.mp3”)
2. a json file (“main_questionnaire.json”) containing the self-reported information related to the COVID-19 test (result, type and date of the test), vaccination against COVID-19 status, COVID-19 related symptoms, vital signs and more.
3. a json file (“breathing_features.json”) containing data from our meta-analysis on the data. Specifically, it contains features, such as respiratory rate and time parameters, estimated from the meta-analysis on the regular breath recordings. It also contains ground truth annotations for the inhale and exhale periods of the regular breath recordings.
4. four json files (“experts.breath.json”, “experts.cough.json”, “experts.medical_advice.json”, “experts.speech.json”) containing experts’ annotations on 4 data labeling campaigns.

Finally, there is also a csv file “smarty4covid_tabular_data.csv”. This file contains any information available for every submission in the dataset.

A detailed description of the various JSON files and their fields is provided in "smarty4covid_overview.pdf" file.

├── ...
├── <ParticipantID>
│      ├── demographics_underlying_conditions.json
│      ├── <SubmissionId1>
│      │      ├── audio.breath_regular.mp3
│      │      ├── audio.breath_deep.mp3
│      │      ├── audio.cough.mp3
│      │      ├── main_questionnaire.json
│      │      ├── breathing_features.json
│      │      ├── experts.breath.json
│      │      ├── experts.cough.json
│      │      ├── experts.medical_advice.json
│      │      └── experts.speech.json
│      ├── ...
│      └── <SubmissionIdN>
│              ├── audio.breath_regular.mp3
│              ├── audio.breath_deep.mp3
│              ├── audio.cough.mp3
│              ├── main_questionnaire.json
│              ├── breathing_features.json
│              ├── experts.breath.json
│              ├── experts.cough.json
│              ├── experts.medical_advice.json
│              └── experts.speech.json
├── ...
├── smarty4covid_tabular_data.csv
├── smarty4covid_overview.pdf
├── smarty-ontology.owl
├── smarty-triples.nt
├── Smarty4Covid experts info.xlsx
└── readme.xlsx