Awesome

WeSpeaker

WeSpeaker mainly focuses on speaker embedding learning, with application to the speaker verification task. We support online feature extraction or loading pre-extracted features in kaldi-format.

Installation

Install python package

pip install git+https://github.com/wenet-e2e/wespeaker.git

Command-line usage (use -h for parameters):

$ wespeaker --task embedding --audio_file audio.wav --output_file embedding.txt
$ wespeaker --task embedding_kaldi --wav_scp wav.scp --output_file /path/to/embedding
$ wespeaker --task similarity --audio_file audio.wav --audio_file2 audio2.wav
$ wespeaker --task diarization --audio_file audio.wav

Python programming usage:

import wespeaker

model = wespeaker.load_model('chinese')
embedding = model.extract_embedding('audio.wav')
utt_names, embeddings = model.extract_embedding_list('wav.scp')
similarity = model.compute_similarity('audio1.wav', 'audio2.wav')
diar_result = model.diarize('audio.wav')

Please refer to python usage for more command line and python programming usage.

Install for development & deployment

Clone this repo

git clone https://github.com/wenet-e2e/wespeaker.git

Create conda env: pytorch version >= 1.12.1 is recommended !!!

conda create -n wespeaker python=3.9
conda activate wespeaker
conda install pytorch=1.12.1 torchaudio=0.12.1 cudatoolkit=11.3 -c pytorch -c conda-forge
pip install -r requirements.txt
pre-commit install  # for clean and tidy code

🔥 News

2024.09.03: Support the SimAM_ResNet and the model pretrained on VoxBlink2, check Pretrained Models for the pretrained model, VoxCeleb Recipe for the super performance, and python usage for the command line usage!
2024.08.30: We support whisper_encoder based frontend and propose the Whisper-PMFA framework, check #356.
2024.08.20: Update diarization recipe for VoxConverse dataset by leveraging umap dimensionality reduction and hdbscan clustering, see #347 and #352.
2024.08.18: Support using ssl pre-trained models as the frontend. The WavLM recipe is also provided, see #344.
2024.05.15: Add support for quality-aware score calibration, see #320.
2024.04.25: Add support for the gemini-dfresnet model, see #291.
2024.04.23: Support MNN inference engine in runtime, see #310.
2024.04.02: Release Wespeaker document with detailed model-training tutorials, introduction of various runtime platforms, etc.
2024.03.04: Support the eres2net-cn-common-200k and campplus-cn-common-200k of damo #281, check python usage for details.
2024.02.05: Support the ERes2Net #272 and Res2Net #273 models.
2023.11.13: Support CLI usage of wespeaker, check python usage for details.
2023.07.18: Support the kaldi-compatible PLDA and unsupervised adaptation, see #186.
2023.07.14: Support the NIST SRE16 recipe, see #177.

Recipes

VoxCeleb: Speaker Verification recipe on the VoxCeleb dataset
- 🔥 UPDATE 2024.05.15: We support score calibration for Voxceleb and achieve better performance!
- 🔥 UPDATE 2023.07.10: We support self-supervised learning recipe on Voxceleb! Achieving 2.627% (ECAPA_TDNN_GLOB_c1024) EER on vox1-O-clean test set without any labels.
- 🔥 UPDATE 2022.10.31: We support deep r-vector up to the 293-layer version! Achieving 0.447%/0.043 EER/mindcf on vox1-O-clean test set
- 🔥 UPDATE 2022.07.19: We apply the same setups as the CNCeleb recipe, and obtain SOTA performance considering the open-source systems
  - EER/minDCF on vox1-O-clean test set are 0.723%/0.069 (ResNet34) and 0.728%/0.099 (ECAPA_TDNN_GLOB_c1024), after LM fine-tuning and AS-Norm
CNCeleb: Speaker Verification recipe on the CnCeleb dataset
- 🔥 UPDATE 2024.05.16: We support score calibration for Cnceleb and achieve better EER.
- 🔥 UPDATE 2022.10.31: 221-layer ResNet achieves 5.655%/0.330 EER/minDCF
- 🔥 UPDATE 2022.07.12: We migrate the winner system of CNSRC 2022 report slides
  - EER/minDCF reduction from 8.426%/0.487 to 6.492%/0.354 after large margin fine-tuning and AS-Norm
NIST SRE16: Speaker Verification recipe for the 2016 NIST Speaker Recognition Evaluation Plan. Similar recipe can be found in Kaldi.
- 🔥 UPDATE 2023.07.14: We support NIST SRE16 recipe. After PLDA adaptation, we achieved 6.608%, 10.01%, and 2.974% EER on trial Pooled, Tagalog, and Cantonese, respectively.
VoxConverse: Diarization recipe on the VoxConverse dataset

Discussion

For Chinese users, you can scan the QR code on the left to follow our offical account of WeNet Community. We also created a WeChat group for better discussion and quicker response. Please scan the QR code on the right to join the chat group.

<img src="https://github.com/wenet-e2e/wenet-contributors/blob/main/wenet_official.jpeg" width="250px">	<img src="https://github.com/wenet-e2e/wenet-contributors/blob/main/wespeaker/wangshuai.jpg" width="250px">

Citations

If you find wespeaker useful, please cite it as

@article{wang2024advancing,
  title={Advancing speaker embedding learning: Wespeaker toolkit for research and production},
  author={Wang, Shuai and Chen, Zhengyang and Han, Bing and Wang, Hongji and Liang, Chengdong and Zhang, Binbin and Xiang, Xu and Ding, Wen and Rohdin, Johan and Silnova, Anna and others},
  journal={Speech Communication},
  volume={162},
  pages={103104},
  year={2024},
  publisher={Elsevier}
}

@inproceedings{wang2023wespeaker,
  title={Wespeaker: A research and production oriented speaker embedding learning toolkit},
  author={Wang, Hongji and Liang, Chengdong and Wang, Shuai and Chen, Zhengyang and Zhang, Binbin and Xiang, Xu and Deng, Yanlei and Qian, Yanmin},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}

Looking for contributors

If you are interested to contribute, feel free to contact @wsstriving or @robin1001