# Attribution

`s2s-a0-en` contains Mimi-tokenized (kyutai/mimi, CC-BY-4.0) derivatives of:

- **Multilingual LibriSpeech (English)** — via `parler-tts/mls_eng`. CC-BY-4.0.
  Pratap et al., "MLS: A Large-Scale Multilingual Dataset for Speech Research" (2020).
- **LibriTTS-R** — via `blabble-io/libritts_r`. CC-BY-4.0.
  Koizumi et al., "LibriTTS-R: A Restored Multi-Speaker Text-to-Speech Corpus" (2023).
- **gemini-flash-2.0-speech** — `shb777/gemini-flash-2.0-speech`. Apache-2.0.

Text streams are tokenized with the Qwen3-0.6B tokenizer (Apache-2.0).
This bucket contains no audio waveforms, only discrete codec tokens + token ids.