Music Information Retrieval at Spotify and the Future of ML Tooling with Andreas Jansson of Replicate
Audio is streamed directly from the publisher (podcasts.captivate.fm) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
Andreas Jansson is the co-founder of Replicate, a version control tool for machine learning. He holds a PhD from City University of London in Music Informatics and was previously a machine learning engineer at Spotify, researching and applying algorithms for music information retrieval.
Learn more about Andreas:
https://www.linkedin.com/in/janssonandreas/
Every Thursday I send out the most useful things I’ve learned, curated specifically for the busy machine learning engineer. Sign up here: http://bitly.com/mle-newsletter
Follow Charlie on Twitter: https://twitter.com/CharlieYouAI
Subscribe to ML Engineered: https://mlengineered.com/listen
Comments? Questions? Submit them here: http://bit.ly/mle-survey
Take the Giving What We Can Pledge: https://www.givingwhatwecan.org/
Timestamps:
02:30 Andreas Jansson
07:30 Overview of music information retrieval (MIR)
13:30 Why use spectrograms and not raw audio?
19:55 The potential for transformers in MIR
22:45 Most exciting applications for ML in MIR
29:20 Challenges in putting ML into production
36:45 What Andreas imagines for the future of ML tools
41:45 Why he's building a tool for ML version control (http://replicate.ai/)
52:55 What Replicate enables via integration or as a platform
01:02:55 Learnings from doing customer discovery for Replicate
01:14:10 "Github for ML models and data"
01:22:30 Rapid fire questions
Links:
WaveNet: a generative model for raw audio
Singing Voice Separation with Deep U-Net CNNs
Joint Singing Voice Separation and F0 Estimation with Deep U-Net Architectures