Fish Audio Drama 3: The Most Emotive Text-to-Speech Model We've Built

Fish Audio has launched Drama 3, a text-to-speech model available through its web platform, mobile apps, and API, designed to deliver more controllable emotional performances from natural-language directions placed in scripts. The company says the model can respond to cues such as whispers, restrained anger, crying, laughter, pacing changes, and emphasis tags, including directions written in languages different from the spoken text, while supporting 83 languages. On Artificial Analysis’ Pronunciation Robustness benchmark, Drama 3 scored 93.2% overall, compared with 88.1% for Fish Audio’s S2.1 Pro, with reported improvements in technical terms, context-dependent words, codes, paths, and email addresses. Drama 3 uses a 35-billion-parameter mixture-of-experts architecture with roughly 3 billion active parameters per token, a long-context hybrid-attention design, and preference-based reinforcement learning intended to reward favored performances rather than discourage expressive delivery. Its weights are available on Hugging Face under the Fish Audio Community License, allowing research, non-commercial use, and qualifying smaller commercial users while prohibiting non-consensual voice cloning. Fish Audio also previewed invite-only interactive editing and audio inpainting tools, and is offering temporary API and creator-plan discounts through November 12.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.