Future gravitational wave experiments such as LISA or Einstein Telescope are expected to observe a variety of overlapped signals on top of a nonstationary noise, GW backgrounds and instrumental artifacts. In this talk I will show how time-frequency techniques provide an useful framework to address this data analysis problem, ensuring robustness to realistic features and increasing speed for the Bayesian inference. I will compare short time Fourier transforms and Wilson-Daubechies domains and, as a practical application, I will talk about the possibility of disentangling stochastic GW backgrounds from noise in a realistic LISA scenario.