The Infinite Drum Machine is a 2016 Google Creative Lab AI experiment organizing 6,330 untagged everyday sounds (inc. Philharmonia Orchestra) via t-SNE dimensionality reduction—placing similar sounds close for exploration/sequencing.
Built by Kyle McDonald, Manny Tan, Yotam Mann; Chrome-optimized, portrait mode best. Free/open-source; part of AI Experiments showcasing unsupervised audio ML.
Core Features 🎧
Interactive sound map: Drag/zoom neighborhoods (e.g., drums near percussive). 16-step sequencer: Assign 4 sounds (kick/snare/hi-hat/perc) to lanes, play loops. Instruments: Tambourine, snare, whip, filter. Tag overlay for discovery; infinite variety via clustering.
No export/save; pure browser play.
Use Cases and Workflows 🥁
Music hobbyists: Explore clusters → Sequence beats → Remix worldly rhythms. Educators: Teach ML/audio viz (t-SNE demo). Producers: Sound hunt inspiration. Workflow: Load → Pan map → Slot sounds to sequencer → Adjust BPM/effects → Jam live.
Fun for viral clips, no account.
Pricing and Plans 💰
Completely free—browser-based experiment, no plans/subscriptions. Open-source GitHub fork/mod freely.
Comparisons 🔄
t-SNE unsupervised > tagged libraries (e.g., Splice); sequencer simple vs DAWs (Ableton). Creative toy vs pro (no export/MIDI); ML viz unique vs static samplers. Beats Loopcloud in discovery fun.
Limits: No save/export, 2016 UI.
Tips and Best Practices 🎛️
Chrome portrait; zoom rural sounds for rarities. Cluster-hop (e.g., speech→instruments). Layer whip/filter for texture. Screenshot beats; record browser audio. Fork GitHub for exports.
Casual creators/educators; quick fun.
Limitations 👥
Browser-only, no mobile opt beyond portrait; dated 2016—no updates. Sound lib fixed. Music explorers/ML curious; not pros.
Tool Type:
AI audio clustering drum machine
Pros:
👍 Free/open-source infinite sounds 🎉
👍 t-SNE unsupervised discovery
👍 Instant beats no tags
👍 Orchestra/real-world samples
👍 Fun/educational ML viz
Cons:
👎 No export/save
👎 2016 dated UI
👎 Browser-only
👎 No BPM fine-tune
Q&A
Q: How does ML organize sounds?
A: t-SNE clusters 6K+ audio unsupervised—similar near (e.g., drums group).
Q: Get started?
A: Open Chrome portrait, pan map, slot to sequencer, play!
Q: Export beats?
A: No native; screen record or GitHub fork/mod.
Q: Sounds source?
A: Everyday + Philharmonia Orchestra; 6330 total.
Q: Mobile/GitHub?
A: Portrait ok; code
Q: Updates 2026?
A: Legacy experiment; timeless but static.
