Music and Machine Learning
summary
Project Magenta demos actual tools and outcomes: TensorFlow and TensorFlow.js models, an 82‑song back catalog manually converted to MIDI, and a year of MusicVAE interpolation that yields endless melodies which are then hand‑culled and arranged into songs that sound like the band. A 20 MB / ~2M‑word lyrics corpus trains a text model; strict rules (no added notes, only transposition/cutting) force human editing and odd vocal phrasing. Working tools shown: MusicVAE interpolation, Drumify Ableton plugin, Piano Genie fruit interface — all used live or on record and requiring significant human curation to succeed.
AI-written summary · about
chapters (7)
- 0:00Start
- 3:27turn it into a drum beat
- 12:08focus on just a single melody from a single song
- 13:59assign any melody to any instrument
- 14:47trained on a corpus of 20 megabytes of text
- 19:36performing the generative material
- 20:21play you the first minute of the final song
related talks
46mMusic Generation with Magenta: Using Machine Learning in
36mMusic and Art with ML
28mMusic and AI in the Browser with TensorFlow js and Magenta js
47mUnleashing your musical creativity with Magenta.js
21mMusical Deep Neural Networks in the Browser
51mThe Merger of Machine Learning, Music and Programming
01h 04mKeynote 3. Interactive generative models for music
52mAI Lofi HipHop - How (Not) To Generate AI Music