kaldi based x-vector trained on Cn-Celeb database.
export KALDI_ROOT=/kaldi/dir
bash run_cnceleb.sh
-
x-vector(cn-celeb) + PLDA(cn-celeb)
- CN-Celeb Eval Core:
- EER: 16.71%
- minDCF(p-target=0.01): 0.7657
- minDCF(p-target=0.001): 0.8823
-
x-vector(voxceleb) + PLDA(cn-celeb)
- CN-Celeb Eval Core:
- EER: 12.43%
- minDCF(p-target=0.01): 0.6064
- minDCF(p-target=0.001): 0.7381
- https://www.danielpovey.com/files/2018_icassp_xvectors.pdf(X-VECTORS)
- https://arxiv.org/pdf/1911.01799.pdf(Cn-Celeb)
- Voxceleb database:http://openslr.org/49/
- 开源voxceleb模型:https://kaldi-asr.org/models/m7
- CN-Celeb database:http://openslr.org/82/