NetEase Youdao's Confucius4-R2T2: Streaming ASR That Never Rewrites Its Output
dr_cintas · x · 2026-11-06
drcintas kicks off a paper walkthrough of NetEase Youdao's Confucius4-R2T2. Streaming ASR has a hidden problem: most models keep rewriting their output, causing flickering subtitles and voice agents reprocessing the same sentence. R2T2 claims to fix this by never rewriting emitted text. Built on Qwen3-ASR, it supports configurable 80ms–2s decoding chunks, append-only output, 200–600ms average latency with no offline accuracy loss, vLLM and transformers backends, context/hotword prompts, and is Apache 2.0 open source. The thread analyzes what actually works and what could be better.
More from Models
- 8 Types of LLMs Powering AI Agents, from GPT Backbones to LAM Action Models — mdancho84 · 2026-11-07
- ChatGPT Images 2.6 vs Nano Banana 2.1: 4B weekly images vs a 50% price cut — Aiden_Tech_Ai · 2026-11-07
- Benchmarks: Kumo, Nori, Chronos-2 vs tuned LightGBM for power prices — nkafr · 2026-11-07
- a16z: AI Apps Shouldn't Price Per Token — Price at the Highest Value Layer — aleximm · 2026-11-07
- Ex-Rockstar Tech Director: Threat of AI Decompilation Ports May Force a Faster GTA VI PC Release — Ars Technica AI · 2026-11-07
- Running quantized Qwen on a 16GB M2 Pro: 77 tok/s prefill at 120k context — SocietyTomorrow · 2026-11-07