Automatic Transcription of Organ Tablature Music Notation with Deep Neural Networks
Description
This dataset contains the trained models and the test data set presented in the paper "Automatic Transcription of Organ Tablature Music Notation with Deep Neural Networks".
In this paper, we present a deep learning approach to automatically transcribe scores from organ tablature music notation to modern music notation. First, our method segments each input image into the corresponding tablature staves and recognizes tablature characters in the resulting partial images using a deep neural network, consisting of a CNN and a RNN part. The results are converted to the Lilypond music notation format, from which a graphical output in modern music notation can be generated.
We provide downloads for two trained models in the Tensorflow2 SavedModel format (https://www.tensorflow.org/guide/saved_model) in the archive files "model_paper.tar" (the original model presented in the paper) and "model_attention.tar" (a refined model that uses an attention mechanism).
A script to load and run the models can be found in our Git-Repository: https://github.com/umr-ds/organ-tablature-ocr.
We also provide a download for our annotated tablature dataset consisting of 2400 staves from two tablature books by Elias Nikolaus Ammerbach in the archive file "data_Ammerbach.tar". This file contains two folders named "img" and "md", respectively. The "img" folder contains the images and annotations grouped into training (train), validation (val) and test subsets. The images are provided as .png-files, each with a corresponding .txt file containing the annotation. The "md" folder contains the annotations condensed in .csv files for the three subsets.
License
Except where otherwised noted, this item's license is described as Attribution 4.0 International