Skip to main content
POST
/
predict
AI-voice detection
curl --request POST \
  --url https://api.aurigin.ai/v0/predict \
  --header 'Content-Type: multipart/form-data' \
  --header 'x-api-key: <api-key>' \
  --form file='@example-file' \
  --form user_id=speaker_123
{
  "predictions": [
    "fake",
    "fake",
    "real"
  ],
  "global_probability": [
    0.9584,
    0.9585,
    0.9123
  ],
  "error": [
    null,
    null,
    null
  ],
  "model": "apollo-4-2025-10-20",
  "processing_time": 1.350719928741455,
  "audio_duration": 69.91
}
You can send audio data in two ways: via multipart form data (direct file upload) or via JSON with a presigned URL.

​
How It Works

The API processes your audio file in 5-second chunks:
  • Minimum duration: 3 seconds (1 chunk)
  • Maximum duration: 4MB file size
  • Chunk size: 5 seconds each
  • Output: One prediction per 5-second chunk
For example:
  • A 12-second audio file → 3 chunks → 3 predictions
  • A 30-second audio file → 6 chunks → 6 predictions

​
Make a Request

curl
curl -X POST "https://aurigin.ai/api-ext/predict" \
  -H "Content-Type: application/json" \
  -H "x-api-key: <YOUR_API_KEY>" \
  -d '{
    "presigned_url": "https://example.com/presigned.wav",
    "user_id": "optional-user-id"
  }'
1

Authenticate

Include your x-api-key header with every request.
2

Choose upload method

Use multipart upload for local files or supply a presigned_url when the file already lives in storage.
3

Inspect chunked results

Review the per-chunk predictions to understand which segments are AI-generated.

​
Response Breakdown

​
Supported Formats

  • WAV, MP3, M4A, FLAC, OGG
  • Mono or stereo
  • Various bitrates and sample rates
CodeDescription
400Invalid input or file too large (4MB max)
403Authentication failed (check x-api-key)
500Internal error or upstream unavailability
  • Cache presigned URLs for their validity window to minimize round trips.
  • Retry with exponential backoff when you receive 500 errors.
  • Monitor confidence scores to spot borderline results and trigger manual review.

Authorizations

​
x-api-key
string
header
required

Body

​
file
file
required
​
user_id
string

Optional user identifier

Response

OK

​
error
(string | null)[]

Error messages for each 5-second chunk (null if successful). Aligns 1:1 with the predictions array.

​
global_probability
number<float>[]

Confidence scores (0.0-1.0) for each prediction, one per 5-second chunk. Aligns 1:1 with the predictions array.

​
predictions
enum<string>[]

AI detection results for each 5-second chunk of the audio. Array length equals the number of 5-second chunks in the audio file.

Available options:
fake,
real