Do this after you already have a cloned or selected voice in the web app. The API needs a voice reference ID, which the dashboard shows for every voice you own or bookmark.

How do you get a Fish Audio API key?

Sign in, open your account settings, and look for the API keys section. Generate a key and copy it once — treat it like a password. Store it in an environment variable rather than pasting it into scripts you might share.

How do you generate speech with Python?

Install the SDK, then make a call with three inputs: text, a voice reference ID, and an output format.

1. pip install fish-audio-sdk (or use plain requests against the REST endpoint).
2. Set FISH_API_KEY in your environment.
3. Load your text, pass the voice ID, and request mp3.
4. Write the returned bytes to a file per script.

A batch loop over 50 short scripts typically finishes in a couple of minutes on the hosted API.

What are the rate limits and costs?

The API meters usage by characters, same as the web app. A request can hold long text, but splitting into paragraph-sized chunks improves stability and lets you retry only the failed piece. Costs scale with characters generated; heavy volumes justify a paid plan over the free cap.

How do you handle long scripts and errors?

Chunk by paragraph, keep a list of finished files, and skip any chunk already on disk. Retry a failed chunk two or three times with a short delay before moving on. Log the voice ID and text length per file so you can trace a bad render later.

When should you use the API over the web app?

Use the API when the work is repetitive and volume is high — podcast pipelines, e-learning narration, localization. Stay in the web app for one-off clips, since tweaking a line there is faster than re-running a script. The API has no live preview, so expect to listen to output files during setup.

Key Takeaways

• The Fish Audio API generates speech from code using an API key and a voice reference ID.
• Install the Python SDK or call the REST endpoint directly.
• Meter usage is by characters, matching the web app.
• Chunk long text and retry only failed pieces.
• Best for batch narration and localization, not one-off tweaks.

Last updated October 2026. Tool details at Fish Audio.

Whether you are making a podcast intro, dubbing a short video, or building a voice for a game, the process is the same. You do not need a GPU — Fish Audio runs in the browser, and there is also a Python SDK and API for automation.

What is Fish Audio?

Fish Audio is a speech synthesis platform known for its open models (the Fish Speech series) and a hosted web app. You can use built-in voices from a shared library or clone your own. It supports multiple languages and outputs MP3/WAV for download.

How do you clone a voice in Fish Audio?

1. Create a free account and open the Voices section.
2. Click Create voice and upload a sample — 10 to 30 seconds, no background noise, no music.
3. Give the voice a name and accept the terms (it must be your own voice or one you have permission to use).
4. Wait for processing, usually under a minute.
5. Select the voice, paste your script in the text box, and hit Generate.

Download the result as MP3. A 500-word script typically renders in a few seconds on the hosted service.

What makes a good voice sample?

Clean recording beats length. A quiet room, one speaker, steady pace, and a plain read-through of 20 seconds works better than five noisy minutes. Avoid reverb, music beds, and clipping. If you can, read a passage that covers different sounds rather than one flat sentence.

Is Fish Audio free?

There is a free tier with a monthly character limit — enough for short clips and testing. Cloned voices on free accounts are typically shared publicly. Paid plans remove the cap, keep voices private, and add commercial-use rights. Check the current pricing page, since tiers change; the paid entry point has been around $10–$20 per month.

What are the limitations?

Emotion control is blunter than in studio tools — you get less fine control over whispered or shouted delivery. Long scripts can drift in tone near the end. And cloned voices raise legal issues: cloning someone without consent is not allowed and can be illegal depending on where you live.

Key Takeaways

• Fish Audio clones a voice from 10–30 seconds of clean audio.
• Free tier has a monthly character cap; cloned voices are public unless you pay.
• Sample quality matters more than sample length.
• Available as a web app, Python SDK and API.
• Only clone voices you own or have written permission to use.

Last updated October 2026. More details on the Fish Audio tool page.