0xMarioNawfal(@RoundtableSpace):Fish Audio has released Drama 3, an open-weight text-to-speech model that accepts performance directions within the text. → The model uses a 35-billion-parameter mixture-of-experts design with about 3 billion parameters active per token → It supports up to 16 speakers in a single text and 83 languages → Fish Audio reports a 93.2% score on Artificial Analysis' pronunciation robustness benchmark → The weights are available on Hugging Face under a Fish Audio license Source:

2026.10.15 04:45

Fish Audio has released Drama 3, an open-weight text-to-speech model that accepts performance directions within the text. → The model uses a 35-billion-parameter mixture-of-experts design with about 3 billion parameters active per token → It supports up to 16 speakers in a single text and 83 languages → Fish Audio reports a 93.2% score on Artificial Analysis' pronunciation robustness benchmark → The weights are available on Hugging Face under a Fish Audio license Source: