Tag
This paper investigates the Mimi codec's semantic tokens, showing they map to phonetic units like quadphones and triphones, and discusses implications for speech tokenization in AI models.