# Coswara: Crowdsourced Breathing, Cough and Voice Recordings for COVID-19 Detection from 2,746 Participants (IISc Bangalore), 16 kHz Parquet

## Facts

- Provider: Indian Institute of Science (IISc) Bangalore, Coswara project; 16 kHz Parquet edition on Hugging Face
- License: CC-BY-4.0 (https://raw.githubusercontent.com/iiscleap/Coswara-Data/master/README.md)
- Size: 62 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/szzs1693/coswara-data
- Access: public, readable without an account
- Category: Health & Biomedical
- Source: https://doi.org/10.1038/s41597-023-02266-0
- Page: https://scigantic.com/archive/b82dc5e4-864b-5f80-bfea-4a19f7d2ce0b
- Keywords: Health & Biomedical, Parquet, Hugging Face

## About

Coswara, the crowdsourced respiratory sound and speech database collected by the Indian Institute of Science (IISc) Bangalore between April 2020 and February 2022 for COVID-19 screening from audio. Each of 2,746 participants from more than 20 countries (most in India) recorded nine sounds: deep and shallow breathing, heavy and shallow cough, sustained vowels /a/, /e/ and /o/, and counting at normal and fast pace, giving 24,718 audio files, with 681 participants COVID-19 positive. Metadata has 36 fields per participant covering COVID-19 status, demographics, symptoms and comorbidities, plus per-recording audio quality ratings. This Hugging Face edition resamples all audio to 16 kHz WAV and stores it as Parquet in two configurations, metadata and audio, with participant-level train, validation and test splits; load with Hugging Face datasets. Uses include cough and breath sound classification, respiratory disease detection, acoustic biomarkers and health speech analysis. The IISc Coswara-Data repository states the data are licensed under CC BY 4.0. Cite the Coswara data descriptor in Scientific Data, 2023 (doi 10.1038/s41597-023-02266-0) and the Interspeech 2020 Coswara paper (doi 10.21437/Interspeech.2020-2768).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("szzs1693/coswara-data")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `audio/test/audio-test-00001.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/b82dc5e4-864b-5f80-bfea-4a19f7d2ce0b
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/b82dc5e4-864b-5f80-bfea-4a19f7d2ce0b
- Over MCP: `get_schema_card` with id `b82dc5e4-864b-5f80-bfea-4a19f7d2ce0b` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Health & Biomedical datasets](https://scigantic.com/topics/health-and-biomedical)
