Describe a voice in plain words.
Pick the take you like, keep it, and get that same voice back from the studio or your own code.
It starts with a sentence
A warm, low male voice in his fifties, calm and unhurried, a slight rasp, close to the microphone.
Nothing recorded yet
Watch the words turn into him.
A warm, low male voice in his fifties, calm and unhurried, a slight rasp, close to the microphone.
You wrote
A warm, low male voice in his fifties, calm and unhurried, a slight rasp, close to the microphone.
One sentence in. One voice out. Press play.
"Good evening. Pull up a chair, the kettle is on, and we have the whole night. Nobody here is in a hurry."
Real recording from the Studio engine, take B. Not edited.
One sentence, a few candidates.
Each take reads the same description with a different seed. Some land close to each other, some are clearly someone else, so you listen before you choose. The one you keep becomes the reference, and everything after is cloned from it.
Then it says anything, as the same person.
- "Flight two forty to Lisbon is now boarding at gate twelve."
- "That page moved. Nothing is broken, I will take you back home."
- "Once there was a lighthouse keeper who counted ships instead of sheep."
Made in Voice Studio from take B, through the same quality check every API request gets. Same words and seed give you the identical file back.
Call it from your code.
curl https://voice.anri.ai/v1/speech \
-H "Authorization: Bearer $VOICE_KEY" \
-d '{"voice_id": "<your voice>",
"text": "Your order shipped this morning.",
"callback_url": "https://you.example/done"}'Also OpenAI-compatible at /v1/audio/speech, and an MCP server for agents.
If you sent that right now
HTTP/1.1 202 Accepted
Location: /v1/speech/…
{
"status": "queued",
"queue": { "position": 1, "ahead": 0,
"starts_in_secs": 0, "ready_in_secs": 30 }
}Every clip takes its place in one line in front of our GPU, and every response says where it stands. Paid plans join nearer the front. Pass a callback_url and we call you, signed, when it is done.
Try it free.
$0/month
Free · 10 minutes of speech and 2 new voices a month, for personal use.
Both engines are open on every plan. Breeze is a research model, so its audio is never for commercial use.
Use it commercially.
$6/month
Starter · 40 minutes, 5 new voices, ahead of Free in line.
$15/month
Creator · 120 minutes, 15 new voices, ahead of Starter.
$39/month
Pro · 400 minutes, 50 new voices, front of the line.
Studio-engine audio (VoxCPM2, Apache-2.0) on a paid plan is yours to use commercially.