Evaluating Fake Music Detection Performance Under Audio Augmentations
This work addresses the challenge of distinguishing human-composed from generated music for audio security and authenticity applications, but it is incremental as it focuses on testing an existing model under new conditions.
The paper tackled the problem of detecting fake music by evaluating the robustness of a state-of-the-art detection model under audio augmentations, finding that performance decreases significantly even with light augmentations.
With the rapid advancement of generative audio models, distinguishing between human-composed and generated music is becoming increasingly challenging. As a response, models for detecting fake music have been proposed. In this work, we explore the robustness of such systems under audio augmentations. To evaluate model generalization, we constructed a dataset consisting of both real and synthetic music generated using several systems. We then apply a range of audio transformations and analyze how they affect classification accuracy. We test the performance of a recent state-of-the-art musical deepfake detection model in the presence of audio augmentations. The performance of the model decreases significantly even with the introduction of light augmentations.