Hybrid deep learning model for multimodal vocal and lung signal analysis in health monitoring

Description

S Revathi, K Mohanasundaram, and Palanichamy Naveen present, in a 2026 paper, a listening-based health monitor that checks voice and lungs at once. Sound offers a cheap, non-invasive window on the body. Their system feeds recordings through a hybrid neural network that mixes pattern-spotting and sequence-tracking layers. Voice and lung streams run in parallel, and an ensemble step merges the verdicts.

The team first converts each sound into a standard spectral fingerprint. On their tests, the hybrid network hit 92 percent accuracy at spotting voice disorders. It reached 98 percent on respiratory disorders. The combined ensemble scored 98 percent on both tasks. The authors present the dual design as a step toward remote monitoring tools that catch disease earlier, without clinics, needles, or radiation.

Catalog record

Source: https://europepmc.org/articles/PMC13328715