Deep learning (DL) in audio signal processing has received much attention in the last four years, and it is still a growing field. Some attempts were made to translate deep networks from image to audio applications, and we discuss some of the issues that arise. In the present paper, we present DreamSound, a creative adaptation of Deep Dream to sound addressed from two approaches: input manipulation, and sonification design. Our chosen model is YAMNet, a pre-trained deep network for sound classification. We present the original Deep Dream activation maximization function in relation to three filter-based creative variations that were implemented. We find that more interesting results are achieved with this filter-based approach, but that there is still room for experimentation.