Music Transcription

AI-powered audio to MIDI conversion

CREPE

Basic Pitch

ONNX Runtime

WebGPU

Music Transcription Tool

This advanced transcription tool converts audio to MIDI using state-of-the-art neural networks. Features include:

Pitch Detection

Accurate pitch detection using CREPE neural network with multiple model sizes

Note Extraction

Convert detected pitches to musical notes with onset and duration detection

MIDI Export

Export transcribed music as standard MIDI files for use in any DAW

Technical Details: This tool uses CREPE (Convolutional Representation for Pitch Estimation) and Basic Pitch models running on ONNX Runtime with WebGPU acceleration. Multiple model sizes available from tiny (1.9MB) to full (85MB) for accuracy vs speed tradeoffs.

Available Models:

• CREPE Tiny - 1.9MB (fastest)
• CREPE Small - 6.2MB
• CREPE Medium - 22MB (balanced)
• CREPE Large - 49MB
• CREPE Full - 85MB (highest accuracy)
• Basic Pitch - 225KB (Spotify's model)

Features

✓ Real-time pitch detection

✓ Multiple model sizes

✓ WebGPU acceleration

✓ MIDI file export

✓ Onset detection

✓ Velocity estimation

✓ Browser-based processing

Coming Soon

We're optimizing the transcription models for browser deployment. The core CREPE and Basic Pitch models are implemented in our worker and will be available for interactive use shortly.