modelscope / 3D-Speaker

A Repository for Single- and Multi-modal Speaker Verification, Speaker Recognition and Speaker Diarization
Apache License 2.0
1.07k stars 93 forks source link

transcription #43

Closed wwfcnu closed 9 months ago

wwfcnu commented 9 months ago

3d_speaker里有些audio clip没有相应的转录文本

GeekOrangeLuYao commented 9 months ago

每一段音频包含多种设备,我们只标注一次,它们的文本是一样的。 数据集中没有提供方言部分的文本标注,录制时我们要求说话人自由发挥,而非设置了脚本让他们用方言朗读。 部分缺少近讲音频的片段我们认为那些只适合给到speaker任务,这部分也未提供转录文本。