metadata

license: afl-3.0
task_categories:
  - noisy_speech_recognition
language:
  - en
tags:
  - read book
pretty_name: NoisyLibriSpeech_MUSAN

Dataset Card for the Noisy LibriSpeech dataset

Dataset Description

Homepage: Coming Soon
Repository: https://huggingface.co/datasets/zhaoyang9425/NoisyLibriSpeechDataset-MUSAN
Paper: Coming Soon =- Point of Contact: zhaoyang9425@gmail.com

Dataset Summary

The noisy speech corpus is constructed by randomly sampling noise clips from the MUSAN noise dataset and adding them to LibriSpeech dataset. The Signal-to-Noise Ratio (SNR) levels are sampled from a uniform distribution in 0 dB, 5 dB, 10 dB, 15 dB, and 20 dB.

Dataset Structure

same structure with LibriSpeech dataset