Unverified50% confidenceCautionExact time
MMR衡量的是机器是否还能识别，而非人类听者是否还能听懂，ASR模型识别行为与真人感知存在系统性差异
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified通过偏斯皮尔曼相关分析，功能负载与MMR呈正相关63% similarUnverified智能触发器引入LLM语义理解后，用户可用自然语言描述意图由模型自主判断触发时机，代价是确定性降低可能误触发或漏判63% similarUnverified真正理解机器学习与仅会调库的区别体现在模型出错时的诊断能力上62% similarUnverified现代ASR模型在解码每个词时会计算概率分布，模型识别越有把握置信度分数越接近1.062% similarUnverified自然感的关键机制依赖端点检测（VAD）与增量解码的协同，使模型能在用户话未说完时开始组织回应62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919769API
curl https://kongchang.com/api/v1/knowledge/claims/919769MCP
get_claim(id=919769)