AraSpot: Arabic Spoken Command Spotting

This is the official implementation of "AraSpot: Arabic Spoken Command Spotting" paper, which achieved SOTA on the ASC dataset 99.59% accuracy on the test set, by introducing the ConformerGRU as shown in the image below, applying online data augmentation, and synthetic data generation.

Dataset

You can find the ASC dataset here

Results

The below image shows the results across different experiments as illustrated in the paper, while the black horizontal line is the best-performing model on the literature.

Try it out

1. Install the requiremnts using the command below

pip install -r requirements.txt

2. Download the model from here, or the best performing model from here

3. Use the function single_predict in predict.py file.