Access models via API
The Crele API is coming soon. Speaker isolation and speech enhancement as batch endpoints first, then real-time target speaker extraction on a stream. Join the waitlist and tell us what you are building.
Send audio, get one voice back
Send a recording and a few seconds of the voice you want, and get that one speaker back. Send one voice, and get it back clean.
Then the stream: enroll a voice, open a connection, and receive the isolated audio and attribution signals as the audio arrives. The rest of your pipeline never hears the room.
const session = crele.stream({ voice: enrolled })
session.on("audio", chunk => {
sink.play(chunk) // your speaker, isolated
})
session.on("speaker", state => {
state.speaking // true
state.confidence // "high"
state.turn // "yours"
})Built for
- Voice agents
- A front of pipeline that knows who the caller is, so the agent stops taking turns from the television.
- Transcription
- One clean voice in, so speaker labels stop swapping mid-call.
- Call analytics
- Agent and customer out of one mono recording, each on their own.
- Meeting tools
- The person you care about, followed through a room full of people.
- Audio hardware
- Far-field and wearable capture, where the real recordings are made.
More than the free tools can do
What the live engine returns, and where it goes.
The two browser tools are a preview. Short clips, one model each, enough to hear whether the separation holds on audio like yours.
The API is where the caps come off, and it reaches models we do not put behind a public tool.
It also reaches the live engine, which has no browser version at all: enroll a voice, open a connection, and the isolated audio and the attribution signals arrive as the audio does.
Start in the browser
You can check the output on your own files today, before there is a key to ask for.
Run your audio through Isolate a Speaker and Enhance a Recording. If the separation holds on the kind of audio you work with, join the waitlist and tell us what you are building.
Questions
- Is there an API today?
- No, it is coming soon. The two free tools on this site are a preview of what the engine does, capped and running in the browser, and they are the way to check the output on your own audio before there is an endpoint to call.
- What will the API do first?
- Batch first: send a recording and a short voice sample, get that one speaker back alone; send one voice, get it back clean, both without the limits the browser tools run under. The real-time stream, with isolated audio and attribution signals arriving as the audio does, comes after.
- What will it cost?
- API pricing is not set yet. The plans for the tools are on the pricing page.
- Does it need a voice sample?
- For extraction, yes: a few seconds of the target speaker talking alone, sent once. Enhancement takes a single-voice recording and needs nothing else.
- What audio can it take?
- The common formats the free tools accept today, audio or the audio track of a video. The tool pages list them. The stream will take audio as it is captured.
- How do I get a key?
- Join the waitlist below, pick "A product I'm building", and say what it is. The more specific you are about the audio you are working with, the sooner we can hand you something that fits it.
Get a key first
The API is coming soon. Join the waitlist, pick “A product I'm building”, and say what it is. We'll write when there's a key for you.