- Improved search results table with detailed metadata display
- Added functionality to preview original MIDI alongside generated motifs
- Implemented automatic selection of the best result with visual confidence indicators
- Enhanced user experience with a "Try Next Result" button for A/B testing
These updates provide users with better insights into MIDI quality and facilitate easier comparisons between original and generated compositions.
Backend Enhancements:
- Add ParsedMIDIInfo with track analysis and quality issues
- New /api/midi/parse endpoint for MIDI metadata extraction
- Enhanced MIDIService with parseMIDI method
MIDI Preview Player:
- Complete MIDIPlayer class with polyphonic playback
- Basic oscillator mapping (bass=square, mid=triangle, high=sine)
- Lookahead scheduling with proper cleanup
- Automatic stopping when MIDI ends
New UI Flow:
1. Search → Results table with metadata
2. Select MIDI → Shows parsed info (duration, tracks, tempo)
3. Dual transport: Preview Original vs Generate Motif
4. Try Next Result button for easy A/B testing
Search Results Table:
- Title, source, confidence bar, duration, track count
- Quality issues display (short duration, few notes, etc.)
- Auto-select best result, manual selection available
- Visual confidence bars and issue warnings
This makes the project feel "real" - you can now:
- See exactly what MIDI was found
- Verify it's the right song via preview
- Compare original vs procedural synthesis
- Easily try different results
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Role Mapping Rewrite:
- Replace naive hard thresholds with scored feature analysis
- Extract comprehensive track features (polyphony, repetition, phrase continuity)
- Add melody vs accompaniment separation logic
- Competitive role allocation with fallback handling
- Support for 6 roles: bass, drone, ostinato, texture, accents, melody
Chord Detection & Polyphony:
- Extract chord events from simultaneous notes (50ms window)
- Add ChordEvent type with multiple pitches
- SynthesisEngine supports both single notes and chords
- Drone/texture layers use chords, others remain monophonic
Feature Extraction:
- medianPitch, pitchRange, noteDensity analysis
- polyphonyRatio calculation via note overlap detection
- repetitionScore using 4-note pattern matching
- phraseContinuity detection via stepwise motion
- register classification (low/mid/high)
This should dramatically improve "sounds like the song" recognition
by using actual musical features instead of arbitrary cutoffs.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Backend (MVP):
- Express API with /search and /fetch endpoints
- BitMidi and Dongrays search adapters with scoring heuristics
- CORS proxy with disk caching and validation
- Quality assessment and file size limits
Frontend Integration:
- Switch to @tonejs/midi parser for reliable MIDI parsing
- MIDIService for backend API communication
- MotifEngine updated to try real MIDI, fallback to synthetic
- Enhanced error handling and logging
Development:
- Concurrent frontend/backend npm scripts
- Basic project documentation
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
b1rdmania