Mapping Music onto Words Douglas Eck Université de Montréal Computer Science BRAMS (Brain Music and Sound) CIRMMT (Center for Interdisciplinary Research in Music Media and Technology) I will summarize results from a multi-year research project in using machine learning to label audio with words like "jazz" or "trumpet" or "chill". I will discuss how we obtain training data by collecting tags from a popular website called last.fm. Signal processing techniques are then used to extract acoustic information from audio files. The tags and acoustic features are in turn used to train machine learning models to predict whether or not to apply a given tag to a given audio segment. In this (mostly) non-technical talk I will explain how we get from digital audio to words. Much of my focus will be on examples which give an idea how well (or not) these models work. Finally I will discuss how similar methods can be used to build sensible acoustic spaces for music generation in domains like automatic film score generation and automatic music generation for video games.