MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111843 A) filed by Amrita Vishwa Vidyapeetham on September 17, 2026, for Speech Emotion Recognition System Fusing Graph Network And Convolutional Features.
Inventors include Kalathil, Anuraj; Sreelatha Surendran, Poorna; and Jayaprakash, Sreedutt Ram.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: The present disclosure provides a speech emotion recognition system (100) and a method (400) for mapping a speech signal onto an emotion class. The system (100) includes a coefficient extraction unit (106) deriving a sequence of mel-frequency cepstral coefficients, a network construction unit (108) linking a pair of coefficients through an edge on satisfaction of a visibility criterion to populate an adjacency matrix (202), a graph feature extraction unit (118) deriving a first feature set, a heatmap generation unit (110) rendering the adjacency matrix (202) into an intensity image, a convolutional network (112) deriving a second feature set, a fusion unit (114) concatenating the first feature set and the second feature set, and a classification network (116) mapping the concatenation onto the emotion class. Unlike a conventional approach, the present disclosure fuses a graph-based descriptor and a convolutional descriptor for a robust emotion recognition across a virtual assistant and a mental health assessment.
Disclaimer: Curated by HT Syndication.
