Skip to main content

Referential Deepfake Detection: start task

POST 
/api/technology/experimental/referential-deepfake-detection
Experimental feature

Please note that Referential Deepfake Detection is an experimental feature.
It is under development and may change in the future.

Start Referential Deepfake Detection task for comparing two media files. The technology enables you to determine whether a voice in audio data is genuine with respect to an authentic reference media file or whether it is likely to be a deepfake.

Referential Deepfake Detection features​

  • Compares a reference media file (genuine speech) against a questioned media file (potentially deepfake).
  • Produces a single comparison score indicating the likelihood that the questioned media is a deepfake relative to the reference.
  • The score is a log-likelihood ratio where higher positive values indicate higher probability of deepfake. For detailed technical information see the high-level Deepfake Detection technology description.
  • Both reference and questioned files can be multi-channel, but in that case a specific channel must be selected for each multi-channel file with a query parameter.
  • Processing can be limited to a specific time segment for each file using query parameters.
  • If there is too little speech in either processed file, the score will be empty (i.e. null).
  • Parameter max_speech_length applies to both the reference and questioned audio files.

Request​

Responses​

Referential Deepfake Detection task was accepted. Follow the Location header to poll for the task state.

Response Headers
    X-Location

    ⚠️ Deprecated - use Location header instead.

    Example: /api/technology/experimental/referential-deepfake-detection/123e4567-e89b-12d3-a456-426614174000
    Location

    A URL the client should poll for task state and result.

    Example: /api/technology/experimental/referential-deepfake-detection/123e4567-e89b-12d3-a456-426614174000