ai4bharat/SeamlessAlign

BhasaAnuvaad: A Speech Translation Dataset for 13 Indian Languages

Overview

BhasaAnuvaad, is the largest Indic-language AST dataset spanning over 44,400 hours of speech and 17M text segments for 13 of 22 scheduled Indian languages and English.

This repository consists of parallel data for Speech Translation from SeamlessAlign, a subset of BhasaAnuvaad.

How to use

The datasets library allows you to load and pre-process your dataset in pure Python, at scale. The dataset can be downloaded and prepared in one call to your local drive by using the load_dataset function.

• Before downloading first follow the following steps:

  1. Gain access to the dataset and get the HF access token from: https://huggingface.co./settings/tokens.
  2. Install dependencies and login HF:
     • Install Python
     • Run pip install librosa soundfile datasets huggingface_hub[cli]
     • Login by huggingface-cli login and paste the HF access token. Check here for details.

For example, to download the (indic2en or en2indic) config, simply specify the corresponding config name (i.e., "indic2en" for Hindi):

from datasets import load_dataset
bhasaanuvaad = load_dataset("ai4bharat/SeamlessAlign", "indic2en", split="hindi")

Using the datasets library, you can also stream the dataset on-the-fly by adding a streaming=True argument to the load_dataset function call. Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather than downloading the entire dataset to disk.

from datasets import load_dataset
bhasaanuvaad = load_dataset("ai4bharat/SeamlessAlign", "indic2en", split="hindi", streaming=True)
print(next(iter(bhasaanuvaad)))

Citation

If you use BhasaAnuvaad in your work, please cite us:

@article{jain2024bhasaanuvaad,
  title   = {BhasaAnuvaad: A Speech Translation Dataset for 14 Indian Languages},
  author  = {Sparsh Jain and Ashwin Sankar and Devilal Choudhary and Dhairya Suman and Nikhil Narasimhan and Mohammed Safi Ur Rahman Khan and Anoop Kunchukuttan and Mitesh M Khapra and Raj Dabre},
  year    = {2024},
  journal = {arXiv preprint arXiv: 2411.04699}
}

License

This dataset is released under the CC BY 4.0.

Contact

For any questions or feedback, please contact:

• Raj Dabre ([email protected])
• Sparsh Jain ([email protected])
• Ashwin Sankar ([email protected])
• Nikhil Narasimhan ([email protected])
• Mohammed Safi Ur Rahman Khan ([email protected])

Please contact us for any copyright concerns.

Links

• GitHub Repository 💻
• Paper 📄
• Hugging Face Dataset 🤗