Unverified50% confidenceFactExact time
Inconsistent handling of text normalization edge cases across different evaluation toolchains can cause WER values for the same ASR system to differ by 2 to 5 percentage points across different evaluation frameworks.
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
语码转换ASR评测：WER为何失灵及替代方案
Related Claims
UnverifiedLocomo评测数据集存在局限性，不同Memory框架使用不同的judgment prompt导致分数缺乏可比性63% similarUnverified标准自回归模型逐Token从左到右生成文本，在时间步t仅依赖之前的输出，缺乏自我审查和回溯机制，导致错误积累效应62% similarUnverifiedChunks that are too large dilute retrieval precision, while chunks that are too small result in incomplete semantics; the chunking strategy directly determines the retrieval quality of a RAG system.61% similarUnverified当前主流评测基准很难客观衡量顶级模型之间的真实差距，benchmark分数参考价值有限61% similarUnverified上下文感知解析层通常采用AST分析、依赖图遍历和语义嵌入等技术将底层错误信息转化为高层语义描述60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/705579API
curl https://kongchang.com/api/v1/knowledge/claims/705579MCP
get_claim(id=705579)