end-to-end

OmniFlatten is a novel GPT-based model that enables real-time, full-duplex spoken dialogue through a multi-stage post-training technique that integrates speech and text without altering the original architecture.