MUMBAI, India, March 13 -- Intellectual Property India has published a patent application (202641024589 A) filed by Srm Institute Of Science And Technology; and Easwari Engineering College, Chennai, Tamil Nadu, on March 2, for 'a hybrid speech and text based deep learning framework for depression screening using spectrograms.'
Inventor(s) include Sai Smrithi S; Srihasya K; Smith Raj A R; and Dr. R. Nagalakshmi.
The application for the patent was published on March 13, under issue no. 11/2026.
According to the abstract released by the Intellectual Property India: "This invention blends speech and text analysis in a single deep learning model that spots depression automatically. It takes raw speech and turns it into Mel spectrograms, which helps the model pick up on subtle sounds that often show up when someone is depressed. Then a recurrent neural network takes over, following how those vocal patterns shift and evolve as time goes on and while all that's happening, a transformer based language model (BERT) analyzes the words themselves, looking at the context and meaning from converted speech-to-text. Basically, the model processes both sound and language at the same time. Most other systems just focus on either speech or text but rarely both. This one does both, so it's tougher and better at understanding what is really going on beneath the surface. It runs smoothly even on basic hardware, but it's built to handle bigger workloads too. Tests show this hybrid, sequence-driven model pushes detection accuracy up to about 85-87%, which is a solid jump over models that only use convolutional layers. In the end, this invention gives you a flexible, automated way to spot early signs of depression."
Disclaimer: Curated by HT Syndication.
