Unverified50% confidenceBenchmarkExact time
本地部署延迟由三段构成：faster-whisper在GPU上STT约100-300ms、7B参数LLM首Token约200-500ms、TTS合成约100-300ms
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026
Sources
Related Claims
Unverified云端STT服务通常耗时200-500ms，本地部署的Whisper等模型在GPU加速下可压缩至100ms以内76% similarUnverifiedTTS首字节延迟约100-300ms70% similarUnverified云端TTS服务的首字节响应时间通常200-500毫秒，本地部署的轻量TTS模型（如Piper、Kokoro-82M）可将延迟压至50毫秒以内，但音质通常略逊69% similarUnverified通过中转访问海外模型的延迟通常为300-800ms69% similarUnverifiedSTT、LLM、TTS三段串行叠加，仅处理延迟就超过650毫秒，不包括服务间网络跳转耗时68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/572547API
curl https://kongchang.com/api/v1/knowledge/claims/572547MCP
get_claim(id=572547)