SITW Models
Pretrained x-vector and i-vector systems trained on augmented VoxCeleb 1 and VoxCeleb 2. The Speakers in the Wild dataset is used to evaluate the system. Each download also contains a PLDA backend for scoring embeddings (either x-vectors or i-vectors).
SITW Xvector System 1a
- Date
- 2018-05-31
- Uploader
- David Snyder
- Recipe
- egs/sitw/v2
- Kaldi Version
- 447e964
- Model Type
- SID, Xvector TDNN, Wideband
- Error Rate
- EER: 3.5%, minDCF(0.01): 0.34
SITW Ivector System
- Date
- 2018-05-31
- Uploader
- David Snyder
- Recipe
- egs/sitw/v1
- Kaldi Version
- 447e964
- Model Type
- SID, Ivector, Wideband
- Error Rate
- EER: 5.7%, minDCF(0.01): 0.46
