This paper addresses the challenging problem of accurately estimating nonlinear audio effects without paired input and output signals. To achieve this, we study an unsupervised probabilistic approach and present a novel method based on a diffusion generative model that utilizes black-box and gray-box models to estimate unknown nonlinear effects. Compared to existing adversarial methods, we analyze the performance of both methods under varying parameter settings of the effect operator and available processed recording length. Experiments on other distortion effects demonstrate that the diffusion-based approach provides more stable results and is less sensitive to data availability, while the adversarial approach excels at estimating more pronounced distortion effects. In conclusion, this study demonstrates the potential of diffusion models for system identification in music technology and contributes to robust unsupervised blind estimation of audio effects.
