Selling Sheet Music 64. Building “Shazam for Sheet Music”: Songscription CEO Andrew Carlins
Sep 2, 2025
Andrew Carlins, CEO and co-founder of Songscription, discusses his mission to revolutionize music education with AI-powered music transcription. He explains the innovative two-step model that converts audio to MIDI and then to sheet music, highlighting the importance of tailored instruments and diverse genres. Andrew addresses ethical issues, artist control, and the impact of his technology on the music industry. He also shares insights on how musicians can leverage AI and engage with the evolving landscape of music transcription.
AI Snips
Chapters
Transcript
Episode notes
Piano Models Lead Due To Data Quality
- Piano models perform best because piano data is abundant and easier to isolate.
- Other instruments like violin and guitar need cleaner solo recordings and more training to reach piano quality.
Balance Training Data For Robust Models
- Train models with diverse genres and validation sets to cover jazz, theater, and noisy recordings.
- Label and balance training and validation data intentionally to reach commercially usable robustness.
Combine Real And Synthetic Training Data
- Training data comes from artist partnerships, business deals, and synthetic augmentation.
- Synthetic variations (ambient noise, backgrounds) increase model robustness without creating new performances.