Root Note Detection
The files in this folder are related to the Root note detection task. The notebook exploits monophonic Irish folk tunes processed data (that can be found in cre_root_detection.csv file) and with help of machine learning models predicts the root note of a tune. Determination of the root note of each piece of music in the corpus under investigation is a key foundational step in FONN. Accurate root note data allows reliable calculation of key-invariant chromatic pitch class sequences, which have been the primary input for our pattern analysis and melodic similarity work.
NOTE: Deliverable 3.3 of the Polifonia project describes the context and research in more detail.
To use the best trained model for root-note prediction tasks, follow the demo notebook ./RootNoteDemo.ipynb.
Prerequisites
This component requires the cre_root_detection.csv. This file contains the processed data for each tune in the Ceol Rince na hÉireann (CRE) corpus. please see: /.root_key_detection/cre_root_detection.csv
In this deliverable, we employed a factorial design experiment for Decision Tree, Random Forest, and Naive Bayes algorithms. We used a comprehensive list of hyperparameters to select the top-performing models. We also conducted experiments using SMOTE to generate a synthetic balance dataset. Finally, evaluation was done on an unseen dataset, and the obtained results are superior to state-of-the-art models.
Following is the summary of the current work. The experiment notebook ./root-note-detection.ipynb reads the Ceol Rince na hÉireann (CRE) corpus CSV file and then performs the following steps:
- 1- Exploratory Data Analysis, such as null value, classes count, correlations, etc.
- 2- Global settings are defined to control feature selection
- 3- Multiple dataset are created for model development and its evaluation
- 4- Minority classes are balanced with help of SMOTE
- 5- Classification report of state-of-the-art models for root note detection are generated for comparison
- 6- Factorial design experimental setup is developed to evaluate different classification algorithms such as Decision Tree, RandomForest, NaiveBayes
- 7- The best models are selected, and finally they are compared with SOA models, and the best model is saved.
The demo notebook ./RootNoteDemo.ipynb shows how to use the best trained model for new prediction tasks.
Attribution
If you use the code in this repository, please cite this software as follow:
@software{danny_diamond_2022_6566379,
author = {Danny Diamond and
Abdul Shahid and
James McDermott},
title = {{polifonia-project/folk\_ngram\_analysis: FONN
v0.5dev}},
month = may,
year = 2022
}
License
This work is licensed under CC BY 4.0, https://creativecommons.org/licenses/by/4.0/