Unverified50% confidenceBenchmarkExact time
主流TTS方案如ElevenLabs、Azure Neural Voice和Google WaveNet的首字节延迟约为150-300毫秒
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/22/2026
First Seen
Valid until: 10/20/2026
Sources
Related Claims
UnverifiedThe end-to-end latency of a real-time AI tutor is composed of three stacked segments: ASR latency (50–500ms), LLM first-token generation latency (100–2000ms), and TTS first audio frame latency (50–300ms).76% 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75% similarUnverified云端 TTS 服务的首字节响应时间通常为 200-500 毫秒，是延迟瓶颈的主要来源75% 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74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/589577API
curl https://kongchang.com/api/v1/knowledge/claims/589577MCP
get_claim(id=589577)