QAMA is a reference-free (non-intrusive) quality metric for digitised music archives. The QAMA model is useful to predict the perceived audio quality of vinyl recordings. The preservation of cultural heritage through audio archives is primarily focused on the process of digitization, which involves curating audio collections from various analog media formats like wax cylinders, vinyl discs, and 78 RPMs. The purpose of QAMA is to improve accessibility and usability of audio archives. For example, it can be used to retrieve the best quality items or to detect low-quality old vinyl recordings in a collection. The model is optimised for real-world recordings.
The QAMA model has been trained with Vinylset which includes real-world recordings of several genres e.g. classical, jazz, electronic. We collected 620 tracks extracted from original vinyl recordings and labelled them with mean opinion score (MOS).
For additional information on QAMA and Vinylset see the ICASSP'23 paper.
For more information on quality assessment for digital audio archives see the JAES paper.
First, clone the repo to your local machine. To use QAMA install virtualenvwrapper and then create the virtual environment:
mkvirtualenv -p python3.9 QAMA
This will create a virtual environment QAMA. Then install the requirements:
pip install -r requirements.txt
Activate the virtual environment:
workon QAMA
QAMA has been evaluated with a 3-fold cross-validation. We provide the 3 models plus a model trained using the full Vinylset.
qama_fold1.ptqama_fold2.ptqama_fold3.ptqama_full.pt
To predict the quality of your music collection, organise files in a directory and run:
python predict.py --data_dir /path/to/dir/collection
The script creates a csv file in prediction_files with date time format DD-MM-YYYY_hh-mm-ss_qama.csv
The csv files includes predictions of the 4 models and the average of the 3 cross-validation models. This is defined with the column Mean CV.
The model qama_full.pt is trained using all the Vinylset tracks. Unlike the cross-validation models, the full model has not been evaluated in the ICASSP paper although it might be more accurate.
Informal tests show aligned results with the cross-validation models.
You can set full_model=False if you do not want to use QAMA full.
QAMA has been evaluated for vinyl degradations and using files around 10 seconds. The model accepts variable input length but it was not evaluated for very long tracks.
If you use QAMA or Vinylset please cite this paper:
A. Ragano, E. Benetos, and A. Hines "Audio Quality Assessment of Vinyl Music Collections using Self-Supervised Learning", in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2023.
The code is licensed under MIT license.