Description
Asquire Corpus is a collection of breath sounds, cough, and sustained phonation of vowels recorded from the mouth of both healthy subjects and individuals diagnosed with asthma. The dataset also includes biodata, subject's response to a standard performa and pulmonary function test (PFT) reports. The data will be used to develop a AI model to predict the status of lung using various speech and non-speech sounds recorded with a smartphone.
Objective
- Determine which phonation is best for predicting the status of lung.
- To develope a reliable AI model to predict status of lung.
- Develop a disgnosis tool for asthma using smartphone.
My Role
By wearing a data engineer hat for the Asquire project, I designed the data pipeline, annotation scheme, metadata structure, and formatting of the dataset. This involved collaborating with pulmonologists to determine data needs, setting up audio recording workflows, devising a biodata questionnaire, and structuring the dataset for easy research use.
I developed a complete ETL pipeline to process audio data, generate for data and annotation quality report, and prepare the dataset for machine learning. My technical contributions were essential to building this resource.
Sample annotation of asquire data on Audacity
Data Collection
The recordings were captured using three types of smartphone, from low-end to high-end devices. The data of healthy subject was croudsourced from the general public, while the data of asthmatic subjects was collected from the Pulmonary Health Science department of St. John Hospital.
Sounds of five breath, five coughs, and seven phonations were recored from a specific distance from the mouth with a smartphone using asquire web-app. For asthmatic subjects, a post recording was made ofter administering a bronchodilator. The post recording was made 15 minutes after the administration of the bronchodilator. The post recording was made to assess the effect of the bronchodilator on the subject's breathing sounds.
Asquire Flyer for data collection
The dataset consists of 150 asthma patients and 150 control subjects, totaling 300 recordings. Each recording is accompanied by corresponding metadata, including demographic information, medical history, and the results of the PFT report.
For further details, refer
- Asquire: Main post for Asquire project
- Asquire Tako: A web-app designed to record vocal breath sound simultaneously from multiple devices.
- Asquire VAD: Automatic validation of asquire data from crowd sourced data.