Techniques for randomized device interaction are provided. A first communication pattern is selected, with at least a degree of randomness, from a plurality of communication patterns, where each of the plurality of communication patterns specifies one or more audio profiles. A first audio profile specified in the first communication pattern is identified. A first portion of audio is extracted from a first audio file with at least a degree of randomness, and the first portion of audio is modified based on the first audio profile. Finally, the first modified portion of audio is outputted by a first device.