Sound Modelling - parametric methods and deep learning...
summary
Compares concrete parametric pipelines — source‑filter models, SMS tools and the WORLD model — with learned generators like WaveNet/Magenta and Melotron, showing code in Python/Jupyter and libraries such as WaveBender and SuperCollider. Claims models trained on thousands of instruments can be conditioned to swap timbre, neural separators (Spleeter/Deezer) recover stems but voice quality degrades, and drum-to‑MIDI extraction works well enough for reuse; demos and code run end‑to‑end.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:10Sound representation tools
- 4:56Analog to digital sound
- 9:34Deep learning music
- 12:29Computer-assisted music
- 12:51Sound physics basics
- 16:28Parametric synthesis models
- 21:17ML for sound optimization
- 23:03Neural models WaveNet GANSynth
- 26:42Recording unmixing techniques
- 28:51Mellotron synthesis
- 29:12Q&A
related talks
51mThe Merger of Machine Learning, Music and Programming
46mMusic Generation with Magenta: Using Machine Learning in
34mModelling emotional nuance of musical performance
16mDemo: Functional Sound Design
21mMusical Deep Neural Networks in the Browser
43mA Composer’s Guide to Creating with Generative Neural Networks
01h 04mKeynote 3. Interactive generative models for music
45mThe Sound of Your Secrets: Teaching Your Model to Spy So You Can Learn to Defend