MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641125355 A) filed by Vellore Institute Of Technology on October 21, 2026, for System And Method For Reliability-Gated, Discrepancy-Attributed Cross-Modal Delivery Scoring In Semantic-Aware Ai Speech Coaching, With Hierarchical Delivery Planning, Importance-Weighted Regeneration, And Rhetorically Grounded Feedback Generation.
Inventors include D. Sumathi; Harish Y; Suresh Krishna K; and Aadithya A.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A speech coaching system and method provide reliability-gated, attribution-routed evaluation of spoken delivery. The system generates, from a semantic representation of a transcript before evaluating corresponding audio, a hierarchical phrase-level delivery plan comprising symbolic delivery actions and calibrated confidence values. Semantic consistency of each delivery action is verified against the corresponding transcript representation, and low-consistency actions are selectively regenerated using a bounded regeneration budget determined according to rhetorical importance. Vocal-performance features including pitch, energy, speaking rate and pauses are extracted from the spoken delivery and compared with the verified delivery plan using a discrepancy weighting based on rhetorical importance, planner confidence and semantic consistency. A discrepancy attribution engine distinguishes plan unreliability from genuine speaker delivery deviation. Plan-unreliability discrepancies are routed to regeneration-budget updating and excluded from speaker personalization, whereas confirmed delivery deviations are routed to prioritized feedback and adaptive personalization. Unresolved consistency failures receivea neutral action and zero confidence, preventing unreliable plan elements from propagating into subsequent evaluation.
Disclaimer: Curated by HT Syndication.
