jaeyeonkim99/BEATs-tokenizer-AS2M-iter1

BEATs base tokenizer (VQ-KD) — AudioSet-2M, iteration 1

Acoustic VQ tokenizer (1024-entry codebook) trained on AudioSet-2M with ESPnet, following BEATs (Table 4). Distilled with the iteration-0 BEATs encoder as teacher; used to generate the discrete targets for iteration-1 encoder pre-training. Produced by the egs2/audioset/ssl1 recipe (external_tokenizer_model).

Citation

@inproceedings{chen2022beats,
  title={BEATs: Audio Pre-Training with Acoustic Tokenizers},
  author={Chen, Sanyuan and Wu, Yu and Wang, Chengyi and Liu, Shujie and
          Tompkins, Daniel and Chen, Zhuo and Wei, Furu},
  booktitle={ICML}, year={2023}
}
@inproceedings{bharadwaj2025openbeats,
  title={OpenBEATs: A Fully Open-Source General-Purpose Audio Encoder},
  author={Bharadwaj, Shikhar and Cornell, Samuele and Choi, Kwanghee and
          Fukayama, Satoru and Shim, Hye-jin and Deshmukh, Soham and Watanabe, Shinji},
  booktitle={WASPAA}, year={2025}
}