Unverified50% confidenceReleaseExact time
NetEase Youdao released the Confucius 4R2T2 streaming speech recognition model, with latency controlled between 200 and 600 milliseconds, adopting an append-only subtitle strategy, with code licensed under Apache 2.0 and open weights available for download.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/22/2026
First Seen
Valid until: 12/21/2026
Sources
Related Entities
Related Claims
Unverified该设备本地ASR模块将语音转录为文字，语义理解与答案生成依赖上传云端的LLM完成，云端推理往返延迟通常在500ms至2秒之间73% similarUnverifiedState-of-the-art real-time speech transcription latency is approximately 200–400 milliseconds, with mainstream industry solutions including OpenAI's Whisper, Google Speech-to-Text, and AssemblyAI.72% similarUnverifiedAlibaba officially upgraded its real-time speech recognition model FunASR Realtime, achieving a first-character latency of only hundreds of milliseconds and accuracy approaching offline levels.71% similarVerifiedOpenAI's ChatGPT Advanced Voice Mode is based on the GPT-4o end-to-end voice architecture, directly mapping audio input to audio output with latency compressed to under 300 milliseconds.71% similarUnverified传统语音AI采用串行管线设计，即VAD→ASR→LLM→TTS四个独立模块依次传递数据，每个环节引入约100-300毫秒延迟，四段叠加总延迟往往超过1秒70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/941072API
curl https://kongchang.com/api/v1/knowledge/claims/941072MCP
get_claim(id=941072)