Voxta docs

Fish Audio

Connect Voxta to Fish Audio's cloud text-to-speech: voices you cloned on fish.audio, with bracketed audio tags for delivery.

Fish Audio is an inexpensive cloud TTS with low-latency modes. Voices are cloned on fish.audio and picked up from your account: Voxta lists the voices you created there, not the local voice library.

Setup

Get an API key and a voice

Sign up at fish.audio, generate an API key under API Keys, and create (clone) at least one voice on the site.

Add to Voxta

Add Modules → Services → Fish Audio → Install, then follow the install steps. Paste the API Key. The check button confirms it works and shows how many voices are on the account.

Settings

SettingNotes
ModelDrama 3 (default), S2.1 Pro, S2 Pro, S1, or S2.1 Pro (free developer tier).
Latency ModeNormal (best quality), Balanced (default) or Low.
Temperature, Top P, SpeedPer voice.

Audio tags

With emotion tags on, the delivery is written as a bracketed tag in front of the line: [sad]. Audio Tags (advanced) lists the words offered to the AI, one per line:

  • sad writes [sad].
  • laughing=decay(excited) writes [laughing] on the line the AI marked, then [excited] until it marks another, so a single laugh does not become every sentence.
  • moaning=repeat([moaning, panting, breathy], 4..9) writes one of its tags every 4 to 9 words.

Allow Unrestricted Audio Tags (on by default) offers them as examples, so the AI can write its own, like [barely holding it together]. Drama 3 reads those as free-form directions. Older models know fewer tags and may ignore one or read it out loud, so turn it off if you use them.

The same tag format drives Fish S2 and Drama 3 when run locally by audio.cpp.

Fish Audio's default concurrency is low, so parallel requests can queue.

On this page