发表机构
NTT, Inc.（日本电报电话公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对停顿标注歧义导致DER失真问题，提出将标准DER精确分解为停顿归因和核心残差两部分，在不改变DER前提下揭示标准指标掩盖的错误来源。
AI 中文摘要
说话人日志评估对停顿标注中的歧义性敏感，这种歧义可能夸大说话人日志错误率（DER）或掩盖真实的模型错误。我们表明，先前用于容忍停顿的说话人日志评估的形态学闭运算会丢弃片段级别的区分。相反，我们提出一种精确的、考虑重叠的分解方法，将标准DER分解为可归因于停顿的组成部分（由与停顿填充兼容的错误组成）和残余的核心组成部分（可作为内在说话人日志错误的代理）。该分解保持DER不变，而可归因于停顿的组成部分和核心组成部分随停顿阈值单调变化并最终饱和。涵盖合成变换、标注不匹配、跨域评估和紧边界说话人日志的实验表明，该分解揭示了标准DER无法显现的错误来源。
英文摘要
Speaker diarization evaluation is sensitive to ambiguity in pause annotation, which can inflate diarization error rate (DER) or obscure genuine model errors. We show that morphological closing, which has been used for pause-tolerant diarization evaluation, discards segment-level distinctions. Instead, we propose an exact, overlap-aware decomposition of standard DER into a pause-attributable component, consisting of errors compatible with pause filling, and a residual core component that can serve as a proxy for intrinsic diarization errors. The decomposition leaves DER unchanged, while the pause-attributable and core components vary monotonically with the pause threshold and eventually saturate. Experiments spanning synthetic transformations, annotation mismatch, cross-domain evaluation, and tight-boundary diarization show that the decomposition reveals error sources not apparent from standard DER.
CommentsCode: https://github.com/nttcslab-sp/der-decomposition
