Fish Audio Launches Drama 3, a Text-to-Speech Model You Can Direct and Edit
Fish Audio has launched Drama 3, a text-to-speech model that lets developers, studios and creators direct a voice performance in plain language instead of simply generating one. According to the company, users can shift emotion mid-sentence, produce multi-character scenes and re-edit individual words without re-rendering a whole clip.
- Fish Audio launched Drama 3, a text-to-speech model directed through plain-language instructions.
- Features include mid-sentence emotion shifts, multi-speaker scenes in one pass, and audio inpainting to edit only a selected segment.
- Phoneme-level pronunciation control covers English, Chinese, Japanese, Korean and more.
- Drama 3 is available now at fish.audio, with API usage 80% off for the first 30 days.
- Enterprises switching from another voice provider get their first three months free.
Directing a voice instead of generating it
Fish Audio describes itself as a fast-growing AI platform for text-to-speech, speech-to-text and voice cloning. With Drama 3, users describe the tone, pacing and character they want in plain language, then refine the performance line by line until it fits.
The company points to applications including audio dramas and interactive media, audiobooks and multilingual dubbing, games and animation, AI companion experiences, and content creation platforms. Its examples: a drama studio can direct a two-character argument as a single scene, an audiobook narrator can move from calm to trembling mid-sentence, and a game team can fix a single word of dialogue without re-rendering the clip.
"Voice AI has gotten very good at generating speech. The next step is directing a performance," said Rissa Cao, co-founder and CEO of Fish Audio. She called Drama 3 the company's "most emotive and controllable model yet" and said it matters most in long-form projects, where one wrong word should not mean regenerating a whole chapter.
Four new ways to shape a performance
Fish Audio lists four capabilities in Drama 3:
- Natural-language direction: describe a performance in plain language and change a sentence mid-line, so it can start calm and turn panicked without a new take.
- Multi-speaker dialogue: generate a whole scene in one pass, with each character keeping its own voice and responding to what the other just said.
- Audio inpainting: select the part of a performance to edit, update the instructions, and the model regenerates only that segment while leaving the surrounding audio intact.
- Long-form and multilingual consistency: keep the same voice and character across long scripts, switch languages mid-scene, and control pronunciation at the phoneme level in English, Chinese, Japanese, Korean and more.
The company says that, unlike models that stitch separate requests together for long-form content, Drama 3 holds one performance across the full script.
Shijia Liao, co-founder and chief scientist, said Drama 3 "adopts a new architecture that allows for better control and instruction following." On the editing feature, he said that when a single word is changed, the model listens to the audio on both sides of it and makes the new word fit, with the same voice, mood and rhythm. He described it as one of the hardest parts of the model to get right.
Company growth and availability
Drama 3 follows a run of launches from Fish Audio, which marked its first anniversary this summer. In recent months the team shipped a speech-to-text model with built-in speaker diarization, Fish Creative (a single workspace for professional content creators), and mobile apps for iOS and Android. According to the company, Drama 3's predecessor, S2.1 Pro, became the most-requested speech model on OpenRouter within three weeks of launch.
Fish Audio also says it ranked No. 1 on Ahrefs' list of the 50 fastest-growing AI companies, with brand search up 740% year over year. It reports growth from zero to $21 million in annual recurring revenue and more than 15 million users. The company, founded in 2025 and based in Palo Alto, California, grew out of the open-source Fish Speech project and lists HeyGen, OpenArt, Vercel, Runway, Retell, LiveKit and Sanas among teams using its platform.
Drama 3 is available now at fish.audio. For the first 30 days, API usage is 80% off, and enterprises switching from another voice provider get their first three months free. Pricing details are at fish.audio/plan.
What it means for creators and developers
The pitch centers on editing rather than regenerating. For teams producing long audiobooks, dubbed content or game dialogue, fixing one line or word without redoing a whole take could reduce rework, as far as Fish Audio's claims hold in practice. Multi-speaker scenes generated in one pass and phoneme-level pronunciation control address other common production needs. The launch discounts give developers and enterprises a short window to test the model against their current voice providers.
