---

Analysis of the audio file over time

The chart below displays the Fake-O-Meter score, computed using a step size of 4 seconds. Higher values indicate greater likelihood of AI-generated content.

Interpreting the results

The NoveXXV model gave an average score of , suggesting a % deepfake probability.

What has been analysed?

Analysis was limited to audio between second 0 and second 120. No metadata or video was analyzed.

On the 'NoveXXV' model

A state-of-the-art machine learning model trained to identify audio deepfakes.

Limitations

These results are experimental. Current models may not generalize well to evolving deepfake techniques. Only the audio is evaluated. For more, see our research paper .