Fish Audio Drama 3 revises AI voices through director-style notes
Fish Audio's new Drama 3 model adds conversational revision for long, multi-character audio and infilling that can replace a single word without disrupting its surroundings.
AT A GLANCE
Key points
- Supports multi-turn director-style voice revision
- Offers precise single-word audio editing
- Uses a 35B-A3B architecture with about 3B active parameters
- Weights are released under the Fish Audio Community License
What is changing?
Fish Audio announced its new Drama 3 text-to-speech model on October 13. It is designed for multi-character scenes, long-form content, voice cloning, and natural-language control. The company describes Drama 3 as a new flagship with a 35B-A3B architecture that activates about 3 billion parameters per token.
Multi-turn director revision can retain the previous take as context and produce a new performance from notes such as “gentler and a bit sadder.” Audio infilling aims to replace a single word while preserving the surrounding timbre and transitions.
The model weights are released under the Fish Audio Community License, while the previous S2-Pro keeps a separate research license. Local deployment, commercial use, and voice-cloning scenarios therefore require separate checks of the license and speaker consent.
Explore the original sources
Source date: 13 October 2026.
Published by AI Radar: 14 Oct 2026, 00:05 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…