The project focuses on developing a speaker diarization system using End-to-End Neural Diarization (EEND) that can handle variable multiple-speaker scenarios, including overlapping speech, aiming to improve accuracy and efficiency. It addresses the limitation of quadratic memory usage in transformer-based EEND encoder-decoder models. The primary goal is to design an EEND model with a linear encoder-decoder based on the RWKV architecture, optimizing memory efficiency while maintaining high performance in complex diarization tasks.
Shamila Jeewantha
I am a Computer Science and Engineering undergraduate at the University of Moratuwa, with a focus on machine learning and integrated computer engineering. I am excited to explore innovative approaches and contribute to advancements in intelligent systems.
Brief Description of Project
Recent News
- All
