• Feature: 80-dim fbank, mean normalization, speed perturb
• Training: lr [0.0001, 0.1], batch_size 256, 4 gpu(Tesla V100), additive angular margin
• Metrics: EER(%), MinDCF

• Train set: 3D-Speaker-train
• Test set: 3D-Speaker-test

Pretrained models are accessible on ModelScope.

• CAM++: speech_campplus_sv_zh-cn_3dspeaker_16k

• 200k labeled speakers: speech_campplus_sv_zh-cn_16k-common

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

# Install modelscope
pip install modelscope
# CAM++ trained on 3D-Speaker
model_id=damo/speech_campplus_sv_zh-cn_3dspeaker_16k
# CAM++ trained on 200k labeled speakers
model_id=damo/speech_campplus_sv_zh-cn_16k-common
# Run inference
python speakerlab/bin/infer_sv.py --model_id $model_id --wavs $wav_path

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

@article{cam++,
  title={CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking},
  author={Hui Wang and Siqi Zheng and Yafeng Chen and Luyao Cheng and Qian Chen},
  booktitle={Interspeech 2023},
  year={2023},
  organization={IEEE}
}