MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202641048593 A) filed by Dr. Bharathi. N. Gopalsamy; Abhiram S P; and H Keerthi Lakshmi on April 16, 2026, for Exploring Speaker Diarization With Embedding Pipelines.
Inventors include Dr. Bharathi. N. Gopalsamy; Abhiram. S. P; and H Keerthi Lakshmi.
The application for the patent was published on June 12, 2026, under issue no. 24/2026.
Abstract: ABSTRACT Speaker diarization, the process of identifying or labeling a multi-speaker audio with speaker labels, is still a challenging task because of overlapping speech, background noise, and the uniqueness and variability of speakers. Most of the current methods are biased towards either end-to-end neural networks or conventional clustering of chosen features, which have limitations in terms of generalization and scalability In this paper, we present a modular diarization system that combines Whisper transcription systems with unique embedding tools like Resemblyzer and Spectral clustering for speaker label assignment. We also explore segmentation and window-based strategies with modifications, such as smaller window sizes and extra conditions for short audio segments, to increase efficiency. Keywords— Segmentation, Resemblyzer, Spectral clustering, Window-based approach.
Disclaimer: Curated by HT Syndication.
