UniST

This dataset contains UniST codec-token training data exported from local metadata and codec results.

We train UniSS with UniST data.

Schema

• id: sample identifier
• transcription: source transcription from metadata text
• translation: qwen_trans, falling back to trans_text
• source_glm, target_glm: GLM token lists
• source_bicodec, target_bicodec: bicodec semantic token lists
• bicodec_global: source bicodec global token list
• dataset_name, src_lang, tgt_lang, split: source metadata
• *_len, duration_ratio: token-count audit fields

Filtering

Rows are exported only when all four token files exist, source and target bicodec global tokens match, metadata is complete, WER is below the configured threshold, and 0.7 <= len(source_glm) / len(target_glm) <= 1.3.

Included Datasets

• commonvoice_en
• commonvoice_cn
• dailytalk
• emilia_en
• emilia_zh
• hq-conversations
• librispeech
• libritts_r
• ncssd_c_en
• ncssd_c_zh
• ncssd_r_en
• ncssd_r_zh
• vctk
• wenetspeech4tts
• magicdata
• gigaspeech
• hifi_tts

Want to Use Raw Audio?

You can decode the glm tokens with glm4-tokenizer or decode the bicodec tokens with BiCodec.

To decode BiCodec, you can follow the code in UniSS Tokenizer. Remember to provide the global tokens to control spkear information.

Citation

If you find our paper and code useful in your research, please consider giving a star and citation.

@misc{cheng2025uniss_s2st,
      title={UniSS: Unified Expressive Speech-to-Speech Translation with Your Voice}, 
      author={Sitong Cheng and Weizhen Bian and Xinsheng Wang and Ruibin Yuan and Jianyi Chen and Shunshun Yin and Yike Guo and Wei Xue},
      year={2025},
      eprint={2509.21144},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2509.21144}, 
}