Skip to content

📚 Supported Datasets in soundKIT

This document lists the datasets officially supported by soundKIT for speech enhancement tasks. These include speech, noise, and reverb datasets commonly used in research and real-world scenarios.

⚠️ Please ensure that you have the right to use and distribute these datasets. Always consult and respect the original dataset licenses.

🗣️ Speech Datasets

train-clean-100

train-clean-360

  • 📁 Path: wavs/LibriSpeech/train-clean-360
  • 📌 Description: Larger clean speech subset with 360 hours.
  • 🔖 License: Same as above.

dev-clean

  • 📁 Path: wavs/LibriSpeech/dev-clean
  • 📌 Description: Development set for LibriSpeech evaluation.
  • 🔖 License: Same as above.

test-clean

  • 📁 Path: wavs/LibriSpeech/test-clean
  • 📌 Description: Official test set for LibriSpeech.
  • 🔖 License: Same as above.

thchs30

  • 📁 Path: wavs/data_thchs30/train, wavs/data_thchs30/dev
  • 📌 Description: A Chinese speech dataset from THCHS-30.
  • 🔖 License: THCHS-30 License

🔉 Noise Datasets

wham_noise

  • 📁 Path: wavs/noise/wham_noise/tr, wavs/noise/wham_noise/cv
  • 📌 Description: Background noise from the WHAM! dataset used for speech separation and enhancement.
  • 🔖 License: WHAM! License (MIT)

FSD50K

  • 📁 Path: wavs/noise/FSD50K/non_speech.csv
  • 📌 Description: Non-speech audio events from FSD50K.
  • 🔖 License: CC BY 4.0

ESC-50-master

  • 📁 Path: wavs/noise/ESC-50-master/non_speech.csv
  • 📌 Description: Environmental sound dataset containing 50 classes.
  • 🔖 License: CC BY 4.0

musan

  • 📁 Path: wavs/noise/musan/music, wavs/noise/musan/noise
  • 📌 Description: Contains music, speech, and noise segments. Used widely in speaker recognition and enhancement tasks.
  • 🔖 License: MUSAN License (Free for research)

🏠 Reverb Datasets

RIRS_NOISES or rirs_noises

  • 📁 Path: wavs/noise/RIRS_NOISES
  • 📌 Description: A large collection of simulated and real room impulse responses.
  • 🔖 License: OpenSLR License

📌 Notes

  • All dataset loaders are registered through @DatasetRegistry.register("dataset_name").
  • You can inspect or modify the registration logic in soundkit/plugins/register_datasets.py.
  • When adding new datasets, always cite and attribute the original sources as required.

🔒 Licensing Reminder

soundKIT does not distribute any of the datasets listed above. Users are responsible for downloading and using the datasets according to their respective licenses.

If you are publishing research or commercializing a system based on these datasets, please check licensing requirements and attribution rules for each one.