Skip to main content
Fish Audio builds expressive speech models that are directed the way an actor is: with written notes in the script. Its Drama 3 model reads plain-language directions in square brackets and changes the performance from that point in the line.
Available in Ollang Workflows as fishAudio — AI Dubbing. See the Text-to-Speech catalog.

Models

Fish Audio documents 83 languages with automatic language detection. Ollang’s fishAudio provider runs Drama 3 by default.

Key capabilities

  • Plain-language direction. Write the delivery into the line, for example [barely holding it together, forcing a smile]. A new bracket partway through a line changes the read from that point.
  • Sound cues and paired tags such as [shaky breath], <emphasis>…</emphasis>, and <whisper>…</whisper>.
  • Directions in any language, so a script can be annotated by the linguist working in the target language.
  • Voice cloning from short reference audio. Fish Audio asks for at least 10 seconds and recommends two or three clips of 15–20 seconds each.
  • Multi-speaker dialogue in a single generation.

Where it fits in a workflow

Choose Fish Audio when the dub has to act: drama, animation, games, trailers, and promotional content where emotion shifts within a scene. It pairs well with a review step where a linguist adds or adjusts direction before final synthesis. For broader language reach on the same kind of content, compare it with ElevenLabs V4.

Trade-offs

  • Newest model line in the catalog, with the least production history.
  • Output quality tracks the direction. Lines without direction are reliable but less distinctive than a directed read.

Reference