Google’s HeAR-powered AI models can now identify early signs of diseases, such as tuberculosis, through cough sounds. While coughs can be irritating, the sounds we make during speech and breathing can provide medical experts with significant health insights. These bioacoustic sounds are commonly used in healthcare to track, diagnose, monitor, and treat various conditions, including TB and chronic obstructive pulmonary disease (COPD).
In this age of AI advancement, researchers at Google have recognized the potential of sound as a health indicator. With high-quality microphones in smartphones, accessing these sounds for research has become more feasible. Google researchers are exploring AI’s ability to gather health information from audio data.
Google recently introduced Health Acoustic Representations (HeAR), a foundational bioacoustic model that enables the creation of AI models that can interpret human sounds to identify disease markers. Trained with 300 million diverse and de-identified dataset samples, with approximately 100 million cough sounds used to train the cough model specifically, HeAR has achieved high performance with less training data than usual. Google reports that HeAR-trained models surpass other models across various tasks and are better at generalizing across different microphones.
Researchers can now use HeAR to develop customized bioacoustic models with less data, setup, and computation requirements. Swaasa®, a product by Indian respiratory health company Salcit Technologies, uses AI to analyze cough sounds for assessing lung health. Swaasa® is now utilizing HeAR’s capabilities to enhance its bioacoustic AI models, particularly aiming to improve early TB detection through cough sounds.
Stay updated with the latest in technology and gaming at TecMania.
