MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641126811 A) filed by Vellore Institute Of Technology on October 24, 2026, for Method For Document-Based Text-To-Speech Generation Via Spatial Parsing And Dependency Evaluation.
Inventors include Sathyaraj R.; Nirmit Jain; Shayla Tyagi; Rajan Hasija; and Arihant Singh Bhadauria.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: ABSTRACT METHOD FOR DOCUMENT-BASED TEXT-TO-SPEECH GENERATION VIA SPATIAL PARSING AND DEPENDENCY EVALUATION A method (200) for generating audio output from an electronic document is provided. In a text-to-speech pipeline having a spatial extraction engine (104), a block type classifier (108), and a text-to-speech module (118), the improvement includes performing dependency parsing by a dependency parsing router (110) on each text block to determine whether the text block contains a root node with a primary verb. Text blocks containing the root node are added to an audio queue (112) as direct play entries (604). Text blocks lacking the root node are added to a language model queue (114) as syntactic fragments (614) with placeholder entries (606) inserted into the audio queue (112). An adjacent context window (616) is retrieved for each syntactic fragment (614), and a generative language model (116) generates expanded text (620). The placeholder entries (606) are replaced with the expanded text (620), and the text-to-speech module (118) synthesizes audio output in sequential order. (Fig. 1)
Disclaimer: Curated by HT Syndication.
