realfolkcode/dcnn-alignment

Dilated CNN for audio-to-score alignment

This is an unofficial implementation (WIP!) of the paper "Structure-Aware Audio-to-Score Alignment using Progressively Dilated Convolutional Neural Networks".

Instead of training on the MSMD dataset as in the original paper, we train on ASAP with synthetic structural augmentations.

TODO

• Calculate cross-similarity
• Add structural augmentations
• Prepare dataset class
• Add model implementations
• Write training pipeline
• Write inference pipeline