Model attribution

Collin is built on open research and openly licensed components.

Miso TTS 8B

Speech model: Miso TTS 8B by Miso Labs (Kamino Learning, Inc.), Modified MIT license.

Mimi neural audio codec

Audio tokenization: the Mimi codec by Kyutai, CC-BY 4.0.

Built with Llama

Text tokenization uses the Llama 3.2 tokenizer by Meta under the Llama 3.2 Community License. Built with Llama.

SilentCipher

Watermarking: SilentCipher by Sony Research (MIT).