AI Voice Detector API | Proofademic Documentation

Upload audio to detect whether the voice is AI-generated or natural, with a per-segment timeline for longer recordings.

POST /api/voice-detector/detect/

Upload an audio file to detect whether the voice is AI-generated or natural. Short files (up to about 6 seconds) return a single overall verdict; longer files additionally return a per-segment timeline showing which parts of the audio sound AI-generated.

Authentication: X-API-Key header with voice_detector scope

Unlike the text endpoints, this endpoint accepts multipart/form-data — the audio file is sent in a form field named file.

Request Parameters

Parameter	Type	Required	Description
`file`	file	Yes	The audio file to analyze, sent as a `multipart/form-data` field

Supported Formats and Limits

Limit	Value
Formats	WAV, FLAC, OGG, MP3, M4A, AAC
Maximum file size	150 MB
Maximum duration	10 minutes (600 seconds)

For long recordings, prefer a compressed format (MP3, M4A, FLAC) — a 10-minute stereo WAV approaches the size limit, while the same audio as MP3 is a fraction of it. Video files are not accepted.

Pricing

10 credits per second of audio, rounded up to the whole second (minimum 1 second). A 10-minute file costs 6,000 credits. The exact amount charged is returned as credits_used in the response. If the detection fails, the credits are refunded automatically.

Audio length	Credits
4.2 seconds	50
30 seconds	300
10 minutes	6,000

Example

Request

curl -X POST https://developer-portal.proofademic.ai/api/voice-detector/detect/ \
  -H "X-API-Key: YOUR_API_KEY" \
  -F "[email protected]"

Response for short audio (200 OK)

Files up to about 6 seconds return a single overall verdict:

{
  "status": "success",
  "verdict": "ai_voice",
  "ai_probability": 0.92,
  "natural_probability": 0.08,
  "duration_seconds": 4.2,
  "segments": [],
  "billed_seconds": 5,
  "credits_used": 50,
  "execution_time": 1.4,
  "credits_remaining": 1950,
  "user": {
    "id": 12345
  }
}

Response for longer audio (200 OK)

Files longer than about 6 seconds are analyzed in segments and additionally return a timeline and statistics:

{
  "status": "success",
  "verdict": "ai_voice",
  "ai_probability": 0.881,
  "natural_probability": 0.119,
  "duration_seconds": 14.2,
  "segments": [
    { "start_time": "0.0s", "end_time": "6.0s", "duration": "6.0s", "verdict": "ai_voice", "ai_probability": 0.91 },
    { "start_time": "6.0s", "end_time": "12.0s", "duration": "6.0s", "verdict": "ai_voice", "ai_probability": 0.852 }
  ],
  "statistics": {
    "total_segments": 3,
    "processed_segments": 2,
    "ai_segments": 2,
    "natural_segments": 0,
    "ai_time_percentage": 100.0
  },
  "billed_seconds": 15,
  "credits_used": 150,
  "execution_time": 4.8,
  "credits_remaining": 1800,
  "user": {
    "id": 12345
  }
}

Response Fields

Field	Type	Description
`status`	string	Always `"success"` for a completed detection
`verdict`	string	`ai_voice`, `natural_voice`, or (rarely) `undetermined`
`ai_probability`	float	Probability (0–1) that the voice is AI-generated
`natural_probability`	float	Probability (0–1) that the voice is natural
`duration_seconds`	float	Duration of the analyzed audio
`segments[]`	array	Per-segment timeline — empty for short audio
`statistics`	object	Segment statistics — present for longer audio only
`billed_seconds`	integer	Whole seconds billed (duration rounded up)
`credits_used`	integer	Credits charged for this request (`billed_seconds` × 10)
`execution_time`	float	Processing time in seconds
`credits_remaining`	integer	Credit balance after this request
`user.id`	integer	Your user ID

Segments

Each entry in segments[] describes one analyzed span of the audio:

Field	Type	Description
`start_time` / `end_time`	string	Position of the segment in the audio, e.g. `"6.0s"`
`duration`	string	Length of the segment
`verdict`	string	`ai_voice` or `natural_voice` for this segment
`ai_probability`	float	AI probability (0–1) for this segment

A trailing remainder shorter than the analysis window may be skipped; skipped spans are counted in statistics.total_segments but not listed in segments[].

Health Check

GET /api/voice-detector/health/

No authentication required. Returns 200 with "status": "healthy" when the detection service is available.

Errors

Status	Code	Meaning
`400`	`invalid_audio`	The file could not be read as audio — corrupt file or unsupported codec
`400`	`audio_too_long`	Over the 10-minute limit — `details` carries `limit_seconds`
`400`	`invalid_payload`	Missing file, unsupported format, or over the 150 MB size limit
`403`	`insufficient_credits`	Not enough credits — `details` carries `credits` and `required`
`503`	`service_unavailable`	Detection temporarily unavailable — safe to retry shortly
`503`	`service_error`	Detection failed — credits were refunded

Failed requests never charge: validation errors are rejected before any credits move, and failures after the charge are refunded automatically.

See Error Handling for the general error format.

AI Voice Detector

On this page