Quantifying the Ease of Playing Song Chords on the Guitar

and the Billboard Playability dataset

Still under construction This repository contains the code and the Billboard Playability dataset associated with the paper "Quantifying the Ease of Playing Song Chords on the Guitar" In this paper, we focus on rhythm guitar, a popular instrument among amateur musicians, and propose a rubric-based "playability" metric to formalize the spectrum of difficulty.

• We propose an interpretable guitar playability metric;
• the Amsterdam Playability dataset which is an extension of the Billboard dataset of 200 playability annotated songs tested for reliability;
• a rule-based baseline for our playability metric.

Contents

1. data/: Contains the Billboard Playability Dataset of 200 songs, along with the playability annotations, and the code for the billboard and custom data dataloaders in pytorch.
2. models/: Includes the code for the LSTM, and GRU models, used to predict the rubric categories automatically.
3. rule_models: Contains the fitted rule-based model pickle files.
4. trained_models: Contains the trained model pth files.
5. utils/: Contains the rule-based model functions, helper functions for the ML models, and the evaluation metrics.
6. main.py: Contains the code for training the models on the billboard playability dataset and evaluating them.
7. custom_data.py: Contains the code for predicting playability scores for new data.
8. rule

Installation

pip install -r requirements.txt

Usage of code

training on the billboard dataset

Train an lstm on predicting the weighted total playability score for all songs in the dataset

python main.py --chord_encoding char --target weighted_total --model lstm --accelerator cpu 

Training on a Single Category other than the default values (see configuration for more info on arguments)

python main.py --batch_size 64 --chord_encoding char --target weighted_total --k_fold 0 --learning_rate 0.002 --model lstm --dropout 0.5 --max_epochs 20 --num_workers 0 --accelerator cpu --file_path ./data/Annotations.csv

Training on All Categories

python main.py --batch_size 64 --chord_encoding char --target all --k_fold 0 --learning_rate 0.002 --model lstm --dropout 0.5 --max_epochs 20 --num_workers 0 --accelerator cpu --file_path ./data/Annotations.csv

predicting playability scores for new data

predicting new values for the data in ./data/ note that the data should be in the same format as the billboard dataset and for the guitardiagram one has to implement a mapping to guitarfingers like A:maj: ['x', 'o', '2:1', '2:2', '2:3', 'o'].

python custom_data.py --chord_encoding char --target weighted_total --model lstm --accelerator cpu --file_path ./data/Annotations.csv --data_path ./data/Annotations.csv

Configuration

Explain the configuration options available to users. Describe each command-line argument or configuration file option and what it does.

--batch_size: Specify the batch size for training. 
    (_default_: 64)
--chord_encoding: Choose the chord encoding method.
    _default_: 'char', 
    options:['char','dotsplit','guitardiagram']).
--target: Specify the target category or 'all' to train on all categories
    _default_: 'weighted_total'   
    options: ['CFP', 'CFD', 'UC', 'RHC', 'CPT', 'BD', 'R', 'weighted_total']
--k_fold: Specify the k-fold index for cross-validation.
--learning_rate: Set the learning rate for training.
--model: Choose the model architecture 
    default: 'lstm'
    options: ['lstm','gru']
--dropout: Set the dropout rate.
--max_epochs: Specify the maximum number of training epochs.
--num_workers: Set the number of data loading workers.
--accelerator: Choose between 'cpu', 'mps, and 'gpu' for hardware acceleration.
--file_path: Specify the path to the playability data file.
    _default_: './data/Annotations.csv'

The rubric

Criterion	Weight	Very difficult (3 points)	Difficult (2 points)	Easy (1 point)	Very Easy (0 points)
Uncommonness of chord	3	A lot of uncommon chords	Some uncommon chords	Few uncommon chords	No uncommon chords
Chord finger positioning	3	Very cramped or very wide fingerspread	Uncomfortable or spread out fingers	Slightly uncomfortable or spread out fingers	Comfortable hand and finger position
Chord fingering difficulty	2	Mostly chords that require four fingers or barre chords	Some chords require four fingers to be played or are barre chords (not A or E)	Most chords require three fingers or are A or E barre chords	Most chords can be played with two or three fingers
Repetitiveness	2	No repeated chord progressions	A few repeated chord progressions	Quite a bit of repetition of chord progressions	A lot of repetition of chord progressions
Right-hand complexity	2	For some chords multiple inner strings are not strummed	For some chords one inner string is not strummed	For some of the chords one or more outer strings are not strummed	For the chords all strings are strummed
Chord progression time	1	Very quick chord transitions	Quick chord transitions	Slow chord transitions	Very slow chord transitions
Beat difficulty (syncopes/ghostnotes)	0	A lot of syncopes or ghostnotes	Some syncopes or ghostnotes	A few syncopes or ghostnotes	No syncopes or ghostnotes

Table: Proposed rubric for human annotators evaluating the difficulty of playing the chords of a song on the guitar. Although the rubric functions acceptably using the raw scores from the table header, it has even better predictive power when weighting the criteria according to the factor in the weight column. Note that the beat difficulty criterion provides so little extra information that we recommend omitting it (i.e., setting its weight to zero).

If you have any questions, please feel free to contact us.

Citation:

@article{velezvasquez2023,
  title   = {Quantifying the Ease of Playing Song Chords on the Guitar},
  author  = {Vélez Vásquez, M.A. and Baelemans, M.C.E. and Driedger, J. and Zuidema, W. and Burgoyne, J.A.},
  booktitle = {Proceedings of the 24th International Society for Music Information Retrieval Conference},
  year    = {2023},
}

example of song formatting, chords and lab file