dmumtaz/ugmqa

The repository consists of database and the code for UGM audio quality assessment

Non-Intrusive Perceptual Audio Quality Assessment for User-Generated Content Using Deep Learning

The repository provides the code for predicting the quality of user generated multimedia audio. The IIT-JMU-UGM Audio Dataset is a repository consisting of audio clips extracted from various sources and degraded by real-world distortions. The quality assessment algorithm first extracts the vital features from the sound track and then employs a stacked GRU architecture for training the model.

Dependencies Needed:

Python
glob
Keras
Librosa
Numpy
Sklearn
Pandas
Os
sys

To obtain the objective score on the the test clip using the pre-trained model:

Download the model_final.json and model_final.h5 files
Run the predict_github.py file.

To obtain the full dataset please email us:

deebhamumtaz@gmail.com

If you use our code please cite us:

D. Mumtaz, V. Jakhetiya, K. Nathwani, B. N. Subudhi and S. C. Guntuku, "Non-Intrusive Perceptual Audio Quality Assessment for User-Generated Content Using Deep Learning," in IEEE Transactions on Industrial Informatics, doi: 10.1109/TII.2021.3139010.