Developer API . REST . v1

The AI Voice Detector API.

POST an audio clip, get a citable probability with model attribution in under half a second. The same model that runs the public detector, behind one endpoint.

JSON in, JSON out~0.48s medianWebhooks99% on clean audio
What it does

Built for production.

The detector, as an endpoint. Same model, same methodology, same verdict you can cite.

One endpoint

POST and done

Send a file or a URL. JSON in, JSON out. No SDK to install, just HTTP.

Same model

Identical to the detector

The API runs the exact model behind the public detector. 99% on clean audio.

Attribution

24+ generators named

When we recognize ElevenLabs, Resemble, PlayHT, OpenAI and others, we name the source.

Latency

Sub-second verdicts

Roughly half a second of compute on a typical clip. Built for newsroom and call-center speed.

Async

Webhooks for bulk

Queue large batches and we post the results back when each job finishes.

Citable

Permanent verdict URL

Every response includes a permanent, APA-style citation URL and a one-way audio fingerprint.

The response

JSON in. A verdict out.

The API returns the same probability the public detector does. The number a developer gets back is the number a court can cite.

200 OK · application/json
{
  "verdict": "likely_synthetic",
  "probability": 0.973,
  "confidence": "high",
  "model": "ElevenLabs v2",
  "methodology": "v3.2",
  "duration_ms": 482,
  "fingerprint": "sha256:9f8a2c…",
  "citation_url": "https://aivoicedetector.com/v/ab12cd"
}
Reference

Every field, explained.

What comes back on a successful call. Nothing hidden, nothing rounded up.

FieldTypeWhat it means
verdictstringlikely_synthetic, likely_authentic, or uncertain.
probabilitynumber0 to 1, how likely the speech was machine-generated. We never round up.
confidencestringhigh, medium, or low. Driven by audio quality and clip length.
modelstringThe recognized generator (e.g. ElevenLabs v2), or "unknown synthesis".
methodologystringThe methodology version used, stamped on every verdict (e.g. v3.2).
duration_msnumberCompute time for the verdict, in milliseconds.
fingerprintstringA one-way hash of the audio. We never store the audio file itself.
citation_urlstringA permanent, APA-citable page for the verdict. Quote it, file it, subpoena it.
Calibration

A probability, not a promise.

The API tells you when it cannot tell. A probability near 50% means we genuinely do not know, usually because the clip is short or heavily compressed. We return the number and the confidence, and we never round up to look certain.

Auth
Bearer API key
Formats
MP3 · WAV · M4A · WebM
Median verdict
0.48s
Audio retention
Deleted in 24h
FAQ . the API

Common questions.

Short answers. Email the team for anything not here.

Which plans include the API?
API access is included on Starter and above. See pricing for monthly detection limits. Enterprise can request higher throughput, a custom retention policy, and a dedicated retraining cadence.
How do I authenticate?
A bearer API key in the Authorization header. Generate and rotate keys from your account. Keep them server-side, never in client code.
What can I send?
An audio file (MP3, WAV, M4A, WebM) as multipart, or a public URL to fetch. The detector reads the audio track, so video files work too.
How do you handle large batches?
Submit jobs asynchronously and register a webhook. We post each verdict back to your endpoint as it completes, so you do not hold open connections.
What are the rate limits?
Limits scale with your plan's monthly detection allowance. If you need a burst or a higher sustained rate, email us and we will raise it.
What happens to my audio?
Deleted within 24 hours. Saved verdicts keep a one-way fingerprint, never the audio file. We never train on your audio unless you explicitly opt in.

Build on the verdict.

Get a key in your account. No card to start. Same model on every tier.

Get an API key Talk to the team