In an effort to tackle the growing challenge of tuberculosis (TB) in India, Salcit Technologies, a startup based in Hyderabad, is teaming up with Google to utilize the tech giant’s innovative Health Acoustic Representations (HeAR) model. This collaboration aims to enhance the early detection of TB by analyzing cough sounds.

According to a recent blog post by Google, the HeAR model is a cutting-edge bioacoustic tool trained on a vast dataset of around 300 million audio recordings, including approximately 100 million cough sounds. This model promises to expand TB screening capabilities throughout India.

Salcit Technologies will integrate HeAR into its existing product, Swaasa, which was introduced in 2020. Swaasa employs AI algorithms to detect abnormalities in lung function.

Unveiled in March 2024, HeAR is crafted to assist researchers in developing models that can listen to and interpret human sounds to identify early signs of various diseases.

Sujay Kakarmath, a product manager at Google Research, emphasized the advantages of sound as a diagnostic tool. “Sound is more accessible than blood tests or imaging. HeAR can detect issues such as chest x-ray findings, tuberculosis, and even Covid through cough analysis,” Kakarmath explained.

He highlighted a vision where, in areas with limited medical resources, a healthcare worker equipped with a machine learning model and a smartphone could collect sound samples and provide preliminary diagnostic insights. HeAR aims to accelerate the discovery of new acoustic biomarkers.

The Stop TB Partnership, a UN initiative focused on eradicating TB by 2030, has endorsed this approach. Zhi Zhen Qin, a digital health expert with the organization, praised HeAR for its potential to revolutionize TB screening with a low-impact, accessible tool.

Google is now inviting researchers to apply for access to the HeAR API to explore its potential further. The tech company continues to support advancements in diagnostic and monitoring tools, with the goal of improving global health outcomes.