The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

LibriVAD Dataset Generation

This repository contains the scripts necessary to generate the LibriVAD dataset, a large-scale, noise-augmented dataset for Voice Activity Detection (VAD) based on the LibriSpeech corpus.

Setup LibriVAD Requirements

Install Python Libraries: Navigate to the LibriVAD directory and install the required dependencies.
```
pip install -r requirements.txt
```
Download Source Data: Run the setup.py script and choose the LibriSpeech mirror closest to your location from [EU, USA, CN]. This will download the necessary splits of the LibriSpeech dataset.

EU Example:
```
python setup.py EU
```
After downloading train-clean-100, dev-clean, and test-clean, the local setup should take a few minutes.

Creating the LibriVAD Dataset

Generate the Dataset: After the setup is complete, run the create_LibriVAD.py script. Choose the size of the generated dataset: "small", "medium", or "large".

The approximate sizes of the final dataset are 15GB, 150GB, and 1.5TB, respectively.

Small Dataset Example:
```
python create_LibriVAD.py small
```
The generation process can take a few minutes to several hours, depending on the chosen size and your machine's performance.

Warning: The initial space requirement for the setup process (before generating the final dataset) is approximately 61GB. Please ensure you have sufficient disk space.

Citation

I. Stylianou, A.K. Sarkar, N. Dawalatabad, J. Glass, and Z.-H. Tan, "LibriVAD: A Scalable Open Dataset with Deep Learning Benchmarks for Voice Activity Detection," arXiv preprint arXiv:2512.17281 (2025).

Downloads last month: 15