Machine Listening

WaveNet, media materialism, and rhythmanalysis

Abstract

Beyond the sub-field of computer science concerned with the extraction of meaningful information from audio data, that term "machine listening" invokes the knotty questions of what it is to listen, what (if anything) separates listening by machines and by humans, and how listening is entangled with the materiality of the voice. The timely emergence of WaveNet is provocative regarding each of these—it is, perhaps more than anything else, a listening machine. Furthermore, it reveals the limits of a media materialist approach to sonicity, as exemplified by Wolfgang Ernst, when it comes to media that are artificially intelligent. As a corrective, I propose Henri Lefebvre’s “rhythmanalysis,” a theory of the everyday which helps to take into account the ambiguities of WaveNet.