Introducing the Fish Audio Drama 3 API on Pixazo API: Direct a Voice Performance in 83 Languages (50% Off at Launch)

Most text-to-speech reads a script. Fish Audio Drama 3 performs it. It is a new expressive text-to-speech model, live now on the Pixazo API, that lets you direct a voice the way you would brief a voice actor — set the emotion in plain language, drop in a sigh or a pause exactly where it belongs, whisper or rush a single phrase, and even stage a full multi-speaker scene in a single request, across 83 languages. And to mark the launch, it is 50% off for a limited time.

What is the Fish Audio Drama 3 API?

Fish Audio Drama 3 is an expressive, performance-focused text-to-speech model from Fish Audio, available through the Pixazo API. Where a standard TTS endpoint turns text into a flat read, Drama 3 takes stage directions. You write the line, add notes on how it should be delivered, and the model acts it out — with emotion, timing, emphasis and multiple characters, all from one block of text. It runs asynchronously: you submit a request, get back a request_id, and poll a status URL until the audio is ready as an MP3 or WAV file.

It is in preview, so results vary between takes and Fish Audio may refine the model during the preview period — which, for creative work, is part of the fun: re-roll a line until the read lands.

What makes Fish Audio Drama 3 different?

Four things set it apart from ordinary text-to-speech:

How do you direct a performance?

Everything — words, directions and tags — goes into one text field. Bracketed directions are performance notes, not words to be spoken. Here is the same model, directed four different ways:

Three tools do the heavy lifting:

How does multi-speaker dialogue work?

Pass an array of up to eight voice ids in reference_id, then tag each line with <|speaker:N|> to say who is talking. The model renders the whole exchange — distinct voices and all — in a single request. You bring the voice ids from the Fish Audio voice library (the id is the last part of a fish.audio/m/<id> link), and you are responsible for holding the rights to any voice you use.

<|speaker:0|> [cheerful] Morning! Big day today.
<|speaker:1|> [grumbling, half asleep] It is far too early for cheerful.
<|speaker:0|> [teasing] Coffee is on. <whisper>I may have made it strong.</whisper>

What parameters does the API take?

Beyond the script itself, a handful of optional parameters tune the output:

Suggested Read: Best AI Audio Generation Models in 2026: A Comparison Guide

What does the Fish Audio Drama 3 API cost?

Drama 3 is billed on the UTF-8 bytes of your text — every direction, tag and speaker marker included — rounded up to whole 1,000-byte blocks. For English, one character is one byte, so a 1,000-character script is one block; non-Latin scripts use more bytes per character. During the launch offer you pay half the standard rate:

In practice that is about two cents for a typical 200-word English paragraph during the offer (normally four), and roughly $10 per million bytes instead of $20. There are no seat fees or minimums — you pay per request, and every take is billed separately. New accounts also get 10% cashback on their first payment.

Suggested Read: How to Lip-Sync a Video to Any Audio with AI

How do you call the Fish Audio Drama 3 API?

Submit your script to the endpoint with your API key. The call returns a request_id and a polling_url; poll that URL until the status is COMPLETED and you get back the audio file URL.

curl -X POST 'https://gateway.pixazo.ai/fish-audio-drama-3/v1/text-to-speech' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "[warm] It has been a long time. [suddenly serious] We need to talk.",
    "format": "mp3",
    "prosody": { "speed": 1.0 }
  }'

{
  "request_id": "fish-audio-drama-3_019d...",
  "polling_url": "https://gateway.pixazo.ai/v2/requests/status/fish-audio-drama-3_019d..."
}
# GET the polling_url until "status": "COMPLETED",
# then read the output URL (an .mp3 or .wav file).

Full request and response details, headers and error codes are in the Fish Audio Drama 3 API documentation.

Suggested Read: Hume AI vs ElevenLabs: Comparing Two Expressive Text-to-Speech Models

What can you build with it?

Expressive, directable, multi-speaker speech opens up a lot:

Suggested Read: Best Voice Cloning API Providers in 2026: Ranked and Compared

How do you get started?

Getting a voice performance out of Fish Audio Drama 3 takes three steps:

You can also read the full model overview and pricing on the Fish Audio Drama 3 model page.

Suggested Read: Best Audio Generation APIs in 2026

Frequently Asked Questions

Q. What is the Fish Audio Drama 3 API?

It is an expressive text-to-speech API, available on Pixazo, that lets you direct a voice performance in plain language – set emotion with bracketed directions, change specific words with paired tags, and render multi-speaker scenes in a single request, across 83 languages.

Q. How much does the Fish Audio Drama 3 API cost?

It is billed on the UTF-8 bytes of your text at $0.02 per 1,000 bytes – and 50% off, $0.01 per 1,000 bytes, during the launch offer. That is about two cents for a 200-word English paragraph while the offer lasts, with no seat fees or minimums.

Q. What languages does Fish Audio Drama 3 support?

It supports 83 languages, and you can switch language mid-sentence for a word or phrase using experimental language tags.

Q. Can it do multiple speakers?

Yes. Pass up to eight voice ids and tag each line with a speaker marker, and the model renders the whole conversation with distinct voices in one request.

Q. How do I direct the voice?

Write performance notes in square brackets (such as warm or suddenly serious), drop sounds and pauses like a sigh or a pause where they happen, and wrap specific words in paired tags to whisper, rush or emphasise them.

Q. What audio formats does it return?

MP3 (audio/mpeg, at 64, 128 or 192 kbps) or WAV (16-bit mono PCM at 44.1 kHz). The API is asynchronous: you submit a request and poll for the finished file.