Creating Intelligent Music Instruments with Machine Learning
summary
Research builds handheld, embodied instruments that run small predictive RNNs (typically 32 units) on a Raspberry Pi or single‑board computer and use mixture density recurrent neural network models to predict next control gestures from minutes of logged performance data. Tools include OSC, Pure Data, Arduino MIDI and a servo‑driven output arm; one device (the MP) physically moves a lever and plays sound locally. Models train in under an hour, often work live, and musicians sometimes prefer unexpected or noise‑trained behaviors over strictly accurate continuations.
AI-written summary · about
chapters (15)
- 0:00Start
- 1:25Creating Intelligent Instruments with Machine Learning
- 2:13How To Use Machine Learning for Creating Music
- 5:09Machine Learning Background
- 6:01Recurrent Neural Network
- 8:21Midi Data
- 14:31Robo Jam
- 15:24Mixture Density Recurrent Neural Network
- 16:21Imps
- 19:20Wekinator for Deep Learning
- 25:16Interaction Modes
- 29:18The Embodied Predictive Musical Instrument
- 32:27Software
- 32:54Arduino Midi
- 39:18Noise Model
related talks
03mPowered by TensorFlow: Creating a custom, machine learning-powered drumming arm
24mCreating an AI Musician with JavaScript
47mExpressive AI-Generated Music With Google's Performance RNN
21mMusical Deep Neural Networks in the Browser
41mMaking Machines that Make Music
38mMaking Art with Artificial Intelligence: Artists in Conversation
01h 04mKeynote 3. Interactive generative models for music
41mMusic and Machine Learning