Guitar String Estimation on GuitarSet via YourMT3

Description

A project developed at IEM Graz during WS24/25. It uses YourMT3 as a base for an algorithm for guitar string estimation on acoustic guitars.

This repository implements a system for automatic recognition and assignment of played guitar strings in audio recordings based on the YourMT3 model for F0-tracking. It is currently WIP and further instructions and restructuring will follow.

Installation

1. Set up Git LFS:

git lfs install

2. Clone the repository:

git clone https://github.com/SimonBuechner/GuitarStringEstimation.git
cd GuitarStringEstimation

3. Install Python 3.11 in a new environment (recommended with conda or venv):

Create virtual environment: `python -m venv .venv`
Install dependencies: `pip install -r requirements.txt`

4. Install dependencies:

pip install -r requirements.txt

5. Install the dataset:

python install_dataset.py

Follow the instructions in the CLI. For this project, only GuitarSet dataset is needed.

6. Extract notes information the dataset:

python extract_dataset.py

Dataset

This project uses the GuitarSet Dataset, which contains audio recordings and annotations of acoustic guitar performances.

Based on

• YourMT3 - A modified Music Transcription Transformer model
• GuitarSet - Dataset for guitar recordings with annotations

GuitarStringEstimation Project Structure

After installation, the project directory should look like this:

GuitarStringEstimation/
├── .venv/                    
├── amt/                      
│   ├── content/              
│   ├── logs/                 
│   └── src/                  
│       ├── f0-tracking_accuracy/ 
│       ├── model/        
│       ├── .../                        # multiple directories from YourMT3 model        
│       ├── GuitarStringEstimator.py 
│       └── ...                         # multiple .py files referred to in Usage Instructions
├── data/                               # Dataset directory
│   ├── guitarset_yourmt3_16k/ 
│   ├── yourmt3_indexes/      
│   └── logs/                 
├── scripts/                            # Additional scripts
├── .gitignore               
├── .gitattributes           
├── LICENSE                  
└── README.md                

Usage Instructions:

Evaluate accuracy of YourMT3

To run the evaluation script of YourMT3, navigate to the appropriate directory:

cd amt/src/f0-tracking_accuracy
python f0-trackingYMT3_eval_controller.py

Note: These are intensive calculations best performed on a GPU-accelerated machine. For debugging, modify the dbg boolean toggle in the main() call in f0-trackingYMT3_eval.py.

Calculate Beta-Distributions

Calculates the inharmonicity coefficient (beta) distributions:

python betaDistributions.py

Perform statistical analysis on Beta-Distributions

Runs statistical tests and generates histograms:

python betaDistributions_statTest.py

Run Guitar String Estimation

Calculates guitar string estimations for each note in GuitarSet:

python guitar_string_estimator.py

Scripts

Found in the scripts/ folder.

Frequency Estimation Algorithms

Frequency estimation algorithm comparison.

python FrequencyEstimationAccuracy_eval.py

Evaluate Beta-Distributions algorithm

Evaluates the algorithm for finding Beta-Distributions on synthetic testsignals. Must be run from Project root.

python betaDistributions_eval.py