search

modelscope / 3D-Speaker

A Repository for Single- and Multi-modal Speaker Verification, Speaker Recognition and Speaker Diarization
Apache License 2.0
1.02k stars 89 forks source link

使用speaker diarization结合视频的DER结果效果比单音频的还要差，请问这可以微调嘛？ #91

Closed Coconut059 closed 4 months ago

Coconut059 commented 4 months ago

在MISP2021数据集上使用speaker diarization，使用cam++模型，audio_only:MISS=23;FA=2.56;SER=9;DER=35;;audio_visual:MISS=23,FA=2.56;SER=15;DER=40; 在eval数据上DER差距更大分别是36%和48%，请问clustering部分可以微调嘛