A common practice when processing music signals with deep learning is to transform the raw waveform input into a time-frequency representation. This pre-processing step allows having less variable and more interpretable input signals. However, along that process, one can limit the model's learning capabilities since potentially useful information (like the phase or high frequencies) is discarded. In order to overcome the potential limitations associated with such pre-processing, researchers have been exploring waveform-level music processing techniques, and many advances have been made with the recent advent of deep learning.
In this tutorial, we introduce three main research areas where waveform-based music processing can have a substantial impact:
1) Classification: waveform-based music classifiers have the potential to simplify production and research pipelines.
2) Source separation: making possible waveform-based music source separation would allow overcoming some historical challenges associated with discarding the phase.
3) Generation: waveform-level music generation would enable, e.g., to directly synthesize expressive music.
Presentation material: https://docs.google.com/presentation/d/1_ezZXDkyhp9USAYMc5oKJCkUrUhBfo-Di8H8IfypGBM/edit?usp=sharing or https://zenodo.org/record/3529714#.XcG4my2ZPpA.
Jongpil Lee received the B.S. degree in electrical engineering from Hanyang University, Seoul, South Korea, in 2015, the M.S. degree, in 2017, from the Graduate School of Culture Technology, Korea Advanced Institute of Science and Technology, Daejeon, South Korea, where he is currently working toward the Ph.D. degree. He interned at Naver Clova Artificial Intelligence Research in the summer of 2017 and at Adobe Audio Research Group in the summer of 2019. His current research interests include machine learning and signal processing applied to audio and music applications.
Jordi Pons is a researcher at Dolby Laboratories. He is finishing a PhD in music technology, large-scale audio collections, and deep learning at the Music Technology Group (Universitat Pompeu Fabra, Barcelona). Previously, he received a MSc in sound and music computing (Universitat Pompeu Fabra, Barcelona), and his BSc was in telecommunications engineering (Universitat Politècnica de Catalunya, Barcelona). He also interned at IRCAM (Paris), at the German Hearing Center (Hannover), at Pandora Radio (USA, Bay Area), and at Telefónica Research (Barcelona).
Sander Dieleman is a Research Scientist at DeepMind in London, UK, where he has worked on the development of AlphaGo and WaveNet. His current research interest is large-scale generative modeling of perceptual signals (images, audio, video). He was previously a PhD student at Ghent University, where he conducted research on feature learning and deep learning techniques for learning hierarchical representations of musical audio signals. In the summer of 2014, he interned at Spotify in New York, where he worked on implementing audio-based music recommendation using deep learning on an industrial scale.