Skip to main content
The Audio Analysis API detects AI-generated or manipulated voice recordings.
It applies advanced deepfake detection models to evaluate the authenticity of speech, identifying indicators of robotic synthesis, splicing, or cloned voices. This feature is essential for fraud prevention, voice verification, and trust & safety use cases where detecting synthetic audio is critical.
The process runs asynchronously — you upload an audio file, receive a job_id, and then poll for the final verdict and confidence score.

​
Endpoint Overview

Method: POST
Path: /api/v4/deepfake/audio/analyze
Base URL: https://api.dataspike.io
Authentication: Include your API key in the header
ds-api-token: <YOUR_API_KEY> Supported formats: .wav, .mp3, .m4a
Maximum file size: 10 MB
Content type: multipart/form-data containing one file field

​
Example Workflow (cURL)

The Audio Analysis API runs asynchronously.
Each analysis consists of two simple steps:

​
1. Submit an audio file for analysis

Send your audio file to the API to start the deepfake detection job. 
curl -X POST "https://api.dataspike.io/api/v4/deepfake/audio/analyze" \
  -H "ds-api-token: $DATASPIKE_API_KEY" \
  -F "file=@voice.m4a"
Response (example): 
{
  "id": "01827ed4-c928-7a3c-9a30-7ab7cc169d11",
  "status": "queued",
  "job_type": "audio"
}
The API returns a job ID immediately. This means the analysis has been queued — processing runs asynchronously on the backend. Keep this id to check progress and retrieve final results.

​
2. Retrieve job results

JOB_ID="01827ed4-c928-7a3c-9a30-7ab7cc169d11"

curl -X GET "https://api.dataspike.io/api/v4/deepfake/job/$JOB_ID" \
  -H "ds-api-token: $DATASPIKE_API_KEY"
Completed response (example): 
{
  "id": "01827ed4-c928-7a3c-9a30-7ab7cc169d11",
  "status": "completed",
  "job_type": "audio",
  "score": 0.87,
  "verdict": "deepfake_likely"
}
When status is completed, the response includes:
FieldDescription
scoreConfidence value (0–1). Higher values indicate stronger deepfake likelihood.
verdictHuman-readable classification — for example deepfake_likely or deepfake_unlikely.
Tip: Use a score threshold aligned with your use case (for example, flag recordings with score > 0.6 for manual review).

​
Typical Flow

  1. Upload an audio file for analysis.
  2. Store the returned job_id.
  3. Poll the job endpoint every few seconds until status becomes completed.
  4. Read score and verdict to make automated or manual decisions in your system.

​
Practical Tips

  • Keep files ≤ 10 MB. If your recordings are larger, compress (e.g., transcode to mono, lower bitrate) before upload.
  • One file per request. Use the file form field.
  • Backoff when polling. Poll every 1–3 seconds with exponential backoff to avoid rate limits.
  • Sandbox testing (optional): https://sandboxapi.dataspike.io with the same paths and headers.
  • Handle failures: If status is failed, inspect message/error_code and consider retrying with cleaner audio.