How Riffd works
Riffd uses AI audio analysis to break songs into individual parts, detect musical properties, and give you the tools to learn and play along. No sheet music required.
The Pipeline
1
Input
Search for any song or upload an audio file directly.
2
Separate
AI isolates vocals, guitar, bass, drums, piano, and other instruments.
3
Analyze
Detect key, BPM, chord progression, and generate tablature from each stem.
4
Explore
Mix stems, read tabs, loop sections, transpose, and learn at your own pace.
Technology
Demucs
Stem Separation
Meta's hybrid transformer model separates audio into 6 individual stems with state-of-the-art quality.
Basic Pitch
Note Detection
Spotify's neural network converts audio to MIDI — pitch, timing, and velocity for every detected note.
Krumhansl-Schmuckler
Key Detection
Pitch-class profiling algorithm that matches note distributions against major and minor key profiles.
Web Audio API
Playback Engine
Browser-native audio engine for synchronized stem playback with real-time volume, mute, solo, and transpose.
Diatonic Template Matching
Chord Analysis
Windowed pitch-class histograms scored against diatonic chord templates to identify harmonic progressions.
Flask + Vanilla JS
Application Stack
Lightweight Python backend with a zero-dependency frontend. Fast iteration, minimal overhead.
Principles
Built for musicians
Every feature is designed for people who play instruments. Not a tech demo — a tool you actually use in your practice room.
Honest about limitations
AI analysis isn't perfect. We show confidence levels and suppress low-quality results rather than guessing. You deserve accurate information.
Ship and iterate
Core features are live. Tab accuracy, drum detection, and chord analysis are actively improving with every update.