ai-zahran/E2E-R

Code for Fine-tuning Self-Supervised Learning Models for End-to-End Pronunciation Scoring

E2E-R

Code for the article Fine-tuning Self-Supervised Learning Models for End-to-End Pronunciation Scoring.

This library includes code for training an end-to-end pronunciation scoring model.

This code was built using SpeechBrain.

To run the experiments, follow the following steps:

1. Install the requirements by running pip install -r requirements.txt
2. (Optional) Install Kaldi. This is only necessary if you would like to run the LSTM scorer experiment.
3. Place the TIMIT and speechocean762 datasets in the desired data directory.
4. Set the variable DATA_DIR in the run.sh file to the path of the data directory.
5. Run the experiments in run.sh file.

N.B.: We suggest running the experiments in the run.sh file by copying the commands from the file and pasting them into the terminal for easier debugging.

Cite as:

@article{zahran2023fine,
  title={Fine-tuning Self-Supervised Learning Models for End-to-End Pronunciation Scoring},
  author={Zahran, Ahmed and Fahmy, Aly and Wassif, Khaled and Bayomi, Hanaa},
  journal={IEEE Access},
  year={2023},
  publisher={IEEE}
}