baselines

Skit-S2I baselines

This directory contains the code for the baselines in the paper Skit-S2I: An Indian Accented Speech to Intent dataset.

Installation

Download the dataset

wget https://speech-to-intent-dataset.s3.ap-south-1.amazonaws.com/speech-to-intent.zip
unzip speech-to-intent.zip

Install the necessary packages

conda env create -f environment.yml

Training the baseline

python trainer_whisper.py

Testing the baseline

python test.py

Results

Baseline results for end-to-end intent classification from speech using Skit-S2I dataset, the benchmark is tracked here.

Citation

@article{rajaa2022skit,
  title={Skit-S2I: An Indian Accented Speech to Intent dataset},
  author={Rajaa, Shangeth and Dalmia, Swaraj and Nethil, Kumarmanas},
  journal={arXiv preprint arXiv:2212.13015},
  year={2022}
}