Upload, then detect
Ask for a signed upload URL, PUT the file, then call detect with the upload ID. The audio goes straight to storage, not through our app.
Developer API . REST . v1
The AI Voice Detector API.
Upload an audio clip and get a verdict and an AI probability score back as JSON. It is a deepfake voice detection API: the same model that runs the public deepfake voice detection, behind one endpoint.
JSON in, JSON outPer-segment scoresBearer auth
What it does
What the API does.
The detector as an API: the same model as the free check, returning the same verdict and score.
Upload flow
Same model
Identical to the detector
The API runs the same model as the free check on the website.
Segments
Per-segment scores
Clips longer than six seconds are scored in segments, each with a start time, end time and score.
Synchronous
One request, one verdict
Detection returns on the same request, so a single clip needs no polling.
Limits
Read your plan limits
Get your plan, limits and current usage from the account and usage endpoints instead of hardcoding a quota.
Errors
Plain error codes
Every response carries a request_id. Errors use clear codes such as quota_exceeded and upload_expired.
The response
Example request and response.
The same model as the free check. You get the verdict, the score, and a score for every segment.
{
"type": "voice",
"status": "completed",
"verdict": "likely_ai",
"verdict_label": "Likely AI",
"ai_score": 0.73,
"ai_score_percent": 73.0,
"duration": 47,
"segment_count": 8,
"max_ai_score": 0.92,
"segments": [
{ "start": 0, "end": 6, "ai_score": 0.81, "verdict": "likely_ai" }
]
}
Reference
Every field, explained.
What comes back on a successful call to /api/v1/detect/voice.
| Field | Type | What it means |
|---|---|---|
| verdict | string | ai_generated, likely_ai, likely_human or human. |
| verdict_label | string | The same verdict for display: AI Generated, Likely AI, Likely Human or Human. |
| ai_score | number | 0 to 1, how likely the speech is AI-generated. 0.86 and above is ai_generated, 0.50 to 0.85 likely_ai, 0.16 to 0.49 likely_human, below 0.16 human. |
| ai_score_percent | number | The same score as a percentage. |
| duration | number | Length of the clip in seconds. |
| segment_count | number | How many segments were scored. Clips of six seconds or less return one. |
| average_ai_score | number | The mean score across segments. |
| max_ai_score | number | The highest segment score, useful for spotting a synthetic insert in a real recording. |
| segments | array | Each segment with start, end, ai_score and verdict. |
| created_at, processed_at | string | ISO 8601 timestamps. |
Calibration
A probability, not a promise.
Scores between 16% and 85% land in the Likely Human and Likely AI tiers, so a borderline clip reads as borderline. Short or heavily compressed clips give less reliable results.
FAQ . the API
Frequently asked questions.
Email support@aivoicedetector.com for anything not here.
Which plans include the API?
API access is included on Starter and above. See pricing for monthly detection limits. Need higher throughput? Email the team.
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 in MP3, WAV, M4A, OGG, FLAC or AAC format, uploaded to a signed URL the API gives you. Maximum file size depends on your plan.
How do I check many files?
Call the upload and detect endpoints once per file. Each detect call is synchronous and uses one detection from your monthly allowance.
What are the rate limits?
Requests are rate limited per API key, and a 429 response carries a Retry-After header. Your monthly allowance comes from your plan; if you need more, email the team.
What happens to my audio?
The free web check analyzes audio in memory and keeps nothing. See the privacy policy for how API uploads and results are stored. We never train on your audio unless you explicitly opt in.