Fish Audio API Text to Speech & Voice Cloning
Call Fish Audio from your own product: text to speech, speech to text and voice cloning, with API access included in the $99 Studio credit pack.
API access is included with the $99 Studio credit pack: you sign in, create a key under Settings → API Keys, and call the same voice engine that powers the workspace — real-time text to speech with emotion control, speech to text, voice cloning from a short sample, and the low-latency streaming a voice agent needs. There is no monthly subscription and no per-seat pricing; calls spend credits.
What can you build with the Fish Audio API?
The API exists for teams that have outgrown the browser workspace. Four workloads cover almost everything customers run in production:
- Generated narration at scale. Long-form audio — audiobooks, courses, training material — produced from scripts rather than recorded in a booth.
- Voice agents and assistants. Real-time speech in and out, with the latency budget that conversational products need.
- Voice cloning workflows. A short reference recording becomes a reusable voice your application can hold constant across releases.
- Transcription pipelines. Speech to text with multispeaker separation and emotion tags, so downstream search or QA tooling sees who said what, and how.
If you are still deciding what the product can do, the text to speech guide and the voice cloning guide describe the underlying behaviour before you write any code.
How do you get an API key?
- Buy the $99 Studio pack. It is the pack that includes API access, plus the fastest generation queue and the highest concurrency for real-time streams.
- Sign in to the workspace with the same account.
- Open Settings → API Keys and choose Create API Key. Give it a title that identifies the environment —
staging-web,voice-agent-prod— so rotation later is unambiguous. - Store the key in your own secret manager. Keys are shown in the workspace with their creation date, and can be edited or deleted at any time.
One key per environment is the habit worth keeping. When a key leaks or a contractor leaves, you delete that single key and nothing else in your account changes.
What does API access cost?
Nothing beyond the pack itself. There is no separate API plan, no monthly platform fee, and no minimum commit.
| Pack | Price | Credits | What it adds |
|---|---|---|---|
| Starter | $9 | 1,000 | Text to speech, cloning and transcription |
| Creator | $29 | 4,000 | Commercial usage rights, priority queue |
| Studio | $99 | 15,000 | API access, highest concurrency, priority support |
Packs are one-time purchases. Credits stay valid for 12 months, so a launch month that burns through a pack does not force you into a subscription you keep paying for during quiet months.
How do credits work for API calls?
One credit covers roughly 1,000 characters of generated speech, which makes the arithmetic easy to plan around:
- A 1,000-character support answer costs about 1 credit.
- A 30-minute narration script of roughly 30,000 characters costs about 30 credits.
- Ten hours of finished audiobook — around 600,000 characters — costs about 600 credits, comfortably inside the Studio pack.
Transcription and cloning spend credits the same way. Because packs do not expire quickly and do not renew, the practical question is not "what is the monthly bill" but "how much audio will this quarter ship", which is a number most teams can estimate from their content calendar.
Latency, streaming and concurrency
Voice agents fail on latency before they fail on voice quality. Two things matter in production:
- Streaming output. Audio has to start playing before the whole paragraph is generated, which is what keeps a conversation feeling live instead of queued.
- Concurrent requests. The Studio pack carries the highest concurrency for real-time streams, so several sessions can run at once without waiting behind each other.
The same engine answers in the workspace, so latency you measure in the browser is representative of what an integration will see. The voice agent guide walks through the parts of a conversational stack — speech to text, the model turn, and speech back — and where each one usually becomes the bottleneck.
Which capabilities does the platform expose?
Four capability areas are available to integrations, matching the tabs in the on-page player:
- Text to speech — real-time generation with emotion tags such as
[whispering]or[excited], natural-language delivery notes, and multispeaker dialogue generated as one piece of audio. - Speech to text — transcription with multispeaker separation and emotion tags, described in the speech to text guide.
- Voice cloning — a reusable voice built from a short reference recording, with consent and rights requirements covered in the cloning guide.
- Voice agents — the streaming and concurrency combination that conversational products depend on.
The workspace is the source of truth for the current request format and limits; check the API Keys page and the guide for the capability you are integrating before you hard-code assumptions into a client library.
Where should you start?
A sensible first week looks like this:
- Buy the Studio pack and create a staging key.
- Reproduce one call for the capability you actually need — usually text to speech.
- Measure end-to-end latency with your own network and audio player, not with a sample clip.
- Only then wire the call into your product, and keep a second key for production.
If the integration is a conversational product rather than a generator, start with the voice agent guide; if it is narration at scale, start with the text to speech guide and plan credits from the character counts above.
Frequently asked questions
How do I get access to the Fish Audio API?
API access is part of the $99 Studio pack. Buy the pack, sign in to the workspace, then open Settings → API Keys and create a key for the environment that will call the API.
What can the API generate?
The same engine that powers the workspace: real-time text to speech with emotion control, speech to text with multispeaker transcripts, voice cloning from a short sample, and the low-latency streaming that voice agents need.
How are API calls billed?
Every call spends credits from your pack. One credit covers about 1,000 characters, the Studio pack includes 15,000 credits, and purchased credits stay valid for 12 months.
Do I need a separate subscription for production traffic?
No. Credit packs are one-time purchases with no renewal. When the pack runs low you buy another one, and the Studio pack also carries the highest concurrency for real-time streams.
Where do I manage or revoke keys?
Settings → API Keys lists every key you created, with create, edit and delete actions, so a leaked key can be rotated without touching the rest of your account.