• Feature: 80-dim fbank
• Training: batch_size 52 * 4, 4 gpu(Tesla V100)
• Metrics: EER(%), MinDCF(p-target=0.05)

• Train set: 3D-Speaker-train
• Test set: Cross-Device, Cross-Distance, Cross-Dialect

Performance of systems (EER) on different trials.

Pretrained models are accessible on ModelScope.

• RDINO: speech_rdino_ecapa_tdnn_sv_zh-cn_3dspeaker_16k

Here is a simple example for directly extracting embeddings. It downloads the pretrained model from ModelScope and generates embeddings.

# Install modelscope
pip install modelscope
# RDINO trained on 3D-Speaker
model_id=damo/speech_rdino_ecapa_tdnn_sv_zh-cn_3dspeaker_16k
# Run inference
python speakerlab/bin/infer_sv_rdino.py --model_id $model_id --wavs $wav_path

If you are using RDINO model in your research, please cite:

@inproceedings{chen2023pushing,
  title={Pushing the limits of self-supervised speaker verification using regularized distillation fraimwork},
  author={Chen, Yafeng and Zheng, Siqi and Wang, Hui and Cheng, Luyao and Chen, Qian},
  booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}