Aletheia

Is this a real voice? That is the question Aletheia answers. Use it through the REST API or the MCP server, with the same API key as the other products.

#What it does

• Checks one recording, longer than 4 seconds and at most 120 seconds, in any of the supported formats.
• Answers with a verdict for the whole recording: real, fake or not_sure.
• Scores consecutive parts of the recording too, so you can see which parts sound AI-generated.
• Runs on its own, or together with an Eidos speaker verification on the same recording.

#Read the result

Example response, shortened:

{
  "request_id": "req_8f14e45fceea167a5a36dedd",
  "operation_id": "op_6f1ed002ab5595859014ebf0951522d9",
  "object": "deepfake.detection",
  "result": {
    "verdict": "real",
    "score": 0.0731,
    "duration_seconds": 6.4,
    "segments": [
      { "start": 0, "end": 4, "score": 0.061 },
      { "start": 2, "end": 6.4, "score": 0.085 }
    ]
  }
}

The full list of fields is in Detect AI-generated speech.

#Not sure

not_sure means the recording cannot be judged reliably. Try a longer or cleaner recording.

#Recording length

The recording must be longer than 4 seconds and at most 120 seconds. A recording of 4 seconds or less is refused with 422 and the error code audio_too_short; one over 120 seconds with audio_too_long. Trim long recordings to the part you want checked.

#Use the verdict well

• A verdict is a strong signal, not proof. Use it with other checks, and have a person review decisions that matter.
• Phone-quality audio (calls, voice messages) gives less reliable results than clear recordings.
• The check is tuned for English speech.

#Quickstart

Create an API key first (see Authentication), then send a recording as the multipart field file:

curl https://platform.detectif.ai/v1/deepfake/detect \
  -H "Authorization: Bearer $DETECTIF_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -F "file=@clip.wav"

import os
import uuid

import requests

with open("clip.wav", "rb") as f:
    resp = requests.post(
        "https://platform.detectif.ai/v1/deepfake/detect",
        headers={
            "Authorization": f"Bearer {os.environ['DETECTIF_API_KEY']}",
            "Idempotency-Key": str(uuid.uuid4()),
        },
        files={"file": f},
        timeout=300,
    )
resp.raise_for_status()
result = resp.json()["result"]
print(result["verdict"], result["score"])

Send the same Idempotency-Key when you retry, and you get the stored result back instead of a second check.

With MCP, connect your assistant and ask:

Check if /path/call.wav is AI-generated.

Which parts of /path/call.wav sound AI-generated?

The assistant calls detect_deepfake with the recording and reads the verdict for you.

Uploaded files are stored for 30 days. Send store_audio=false (in MCP, the store_audio argument) to not store a file; see Privacy.

#After the check

• Fetch the result again. Results are kept for 30 days. Get a result (GET /v1/operations/{operation_id}, or the get_operation tool) returns it by its operation_id, for example when your connection dropped before the response arrived. It is shared with Eidos, so it is listed under Account.
• Tell us whether it was right. Say whether a result was right sends correct, and optionally what the recording really is (real or fake) and a short note. Your answer is stored with the result, not the audio.

#API reference

Base URL: https://platform.detectif.ai.

#MCP tools