知识卡片
词错误率用于识别而非语音合成
WER以替换、删除、插入总数除以参考词数,包含插入时可超过100%。它衡量转写误差,不是声音好听程度;书中标题写TTS及把wav2vec当TTS应区分,模型比较需固定语种、数据和预处理。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
Fortunately, evaluating the accuracy of STT software is a bit more straightforward. Using an open source STT benchmark dataset, you can count the number of words correctly transcribed. The following listing shows the word error rate (WER) on several benchmark datasets.
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第12章《对话引擎与会话设计》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-12.html)
- 结论依据:摘录中关于词错误率用于识别而非语音合成的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Fortunately, evaluating the accuracy of STT software is a bit more straightforward. Using an open source STT benchmark dataset, you can count the number of words correctly transcribed. The following listing shows the word error rate (WER) on several benchmark datasets.