ElevenLabs has two different tools for identifying AI-generated audio, and they work in fundamentally different ways:
- The AI Speech Classifier is a legacy tool that uses a trained machine learning classifier to predict whether a given audio clip was generated by ElevenLabs. It only works with content generated by older ElevenLabs TTS models.
- The SynthID-based Audio Detector works differently: it scans the audio for a watermark that was cryptographically embedded into it at the time of generation. This makes it significantly more robust and reliable, but because watermarking is new, it can only detect audio generated after watermarking was introduced.
To give users the broadest coverage, we have combined the two into a single ElevenLabs Audio Detector tool, which will check for a SynthID watermark first and fall back to the AI Speech Classifier if no watermark is found.
You'll need to be signed in to use this tool, but it is available on all plans, including free. To access this tool, open your account profile menu in the top right corner of the website, and select ElevenLabs Audio Detector.
For more information, see What is SynthID, and why is ElevenLabs using it?