MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641072390 A) filed by Sr University on June 11, 2026, for A Graph-Based Cross-Attention System For Emotion Recognition Using Facial Micro-Expressions And Speech Prosody.
Inventors include Gurram Shireesha; Dr. Johnson Kolluri; and Dr. Sampath Korra.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: A GRAPH-BASED CROSS-ATTENTION SYSTEM FOR EMOTION RECOGNITION USING FACIAL MICRO-EXPRESSIONS AND SPEECH PROSODY The invention relates to a graph based cross attention system (100) for emotion recognition that integrates facial micro expression analysis (200) and speech prosody modeling (300) into a unified multimodal framework. Facial features are extracted using spatio temporal algorithms such as optical flow and facial landmark graphs, represented as nodes (231) and edges (241) in a facial graph (240). Speech features including pitch, energy, and rhythm are modeled as nodes (331) and edges (341) in a speech graph (340). A Cross Attention Fusion Module (500) dynamically exchanges information between modalities through attention heads (510) and weights (511), producing a fused representation (540). A Graph Neural Network Reasoning Module (600) performs relational inference on the multimodal graph (410) to classify emotions such as happy, sad, angry, surprised, fearful, and neutral. The system achieves high accuracy and interpretability by explicitly modeling inter modal dependencies and supports real time emotion recognition for healthcare, human computer interaction, and intelligent systems.
Disclaimer: Curated by HT Syndication.
