Fish Audio ships Drama 3 TTS: direct voices in plain language, edit single words
lmoroney · x · 2026-10-15
Fish Audio launched Drama 3, a text-to-speech model built for directing performances rather than just generating speech:
- Creators describe tone, pacing, and character in plain language, then refine line by line; emotion can shift mid-sentence (e.g., calm to trembling narration).
- Multi-speaker scenes generate in one pass with voices responding to each other.
- Any single word or segment can be fixed without re-rendering the clip — useful for audiobook corrections and game dialogue edits.
- Target use cases: audio dramas, audiobooks and multilingual dubbing, games/animation, AI companions, and content platforms. CEO Rissa Cao calls it their most emotive and controllable model yet.
Related event: Fish Audio open-sources Drama 3 TTS model, tops leaderboard(27 posts)→
More from Multimodal
- Physical-Modeling Reverb Built with Gemini 4: Shape It by Stretching Boxes, Export IR to DAW — teropa · 2026-10-15
- AssetFurnace: Free Open-Source Local Image-to-3D Pipeline for Game Characters — chongdashu · 2026-10-15
- BaoCut's open-source approach to AI bilingual subtitle alignment, in five pitfalls — dotey · 2026-10-15
- NormalCrafter ComfyUI nodes turn video into temporally stable normal maps — listopalafoto · 2026-10-15
- Grok is testing a Voice Creator: describe a voice in text, pick from generated options — nima_owji · 2026-10-15
- AI-generated label gag: Rawkhaus Records rejects Kelly Clarkson over 'avian-related concerns' — AIandDesign · 2026-10-15