PolySpeech-100-v1 Dataset

PolySpeech-100-v1 is a large-scale multilingual speech dataset encompassing 110 languages.

Repository Structure

• lang=*.parquet: 110 Parquet files, each corresponding to a specific language subset.
• restore_from_parquet.py: A lightweight standalone Python script provided to easily reconstruct the original dataset hierarchy.

Step 1: Clone the repository (or download specific parquet files)

git lfs install
git clone https://huggingface.co/datasets/youngseng/PolySpeech-100-v1
cd PolySpeech-100-v1

Step 2: Install dependencies

pip install pandas pyarrow tqdm

Step 3: Run the restoration script

python restore_from_parquet.py --parquet_dir . --output_dir ./Restored-PolySpeech