← All model drops
Live drop
Drama 3
Fish AudioDropped Oct 13, 2026
35B open-weights MoE TTS that takes plain-language acting directions, shifts emotion mid-sentence, and handles multi-character scenes.
About this release
Live drop
Fish AudioDropped Oct 13, 2026
35B open-weights MoE TTS that takes plain-language acting directions, shifts emotion mid-sentence, and handles multi-character scenes.
About this release
Drama 3 is Fish Audio's 35B-parameter mixture-of-experts text-to-speech model (approximately 3B active per token) that accepts natural-language performance directions alongside the script. It supports voice cloning from references, multi-speaker dialogue in one pass, mid-sentence emotion or style shifts, precise infill editing of single words or phrases, and streaming/long-form synthesis across 83 languages. Weights are released under the Fish Audio Community License on Hugging Face with SGLang-Omni inference support.
Read the official Fish Audio launch pageFirst builds incoming — this lane fills in automatically as people start building with Drama 3. Know one? Submit the URL.