MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202541097068 A) filed by National Institute Of Technology Calicut An Indian on October 08, 2025, for Speech Waveform Artifact Rectification Architecture.
Inventors include Vidhya Kamakshi; Jay Prakash; Sooryakiran B; Mohsina Bilal; Pranshu Jain; Nilesh Mishra; Venkata Krishna Sravya Ramachandrula; Aswani Raj N P; and Seema Yadav.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: ABSTRACT SPEECH WAVEFORM ARTIFACT RECTIFICATION ARCHITECTURE FOR REAL-TIME STUTTER CORRECTION The present invention provides a system for real-time speech disfluency correction, comprising: a processor configured to receive a raw audio waveform representing stuttered speech; an acoustic encoder, a vector quantizer. There is a disfluency detector that applies frame-level triplet loss to identify disfluent segments and generate a mask; a generative adversarial network (GAN) that corrects the DAUs by inpainting masked positions to produce fluent DAUs. The system also has a Transformer-based decoder and a speaker embedding extractor that derives d-vectors from reference speech; and a speaker-conditioned neural vocoder synthesizes a fluent waveform from the mel- spectrograms, preserving speaker identity. The system operates entirely in the audio domain without intermediate text representations, achieving end-to-end latency. A method is also disclosed. FIG. 1
Disclaimer: Curated by HT Syndication.
