philgzl/dns5

DNS5 Challenge data

This is a mirror of the DNS5 Challenge data. The original files were converted from WAV to Opus to reduce the size and accelerate streaming.

⚠️ Only the LibriVox, AudioSet, Freesound, OpenSLR26, and OpenSLR28 data is included. The VCTK, VocalSet, CREMA-D, VoxCeleb2, and DEMAND data is excluded. ⚠️

• —Sampling rate: 48 kHz
• —Channels: 1
• —Format: Opus
• —Splits:
• —speech_english: 245 hours, 186743 files
• —speech_french: 95 hours, 60454 files
• —speech_german: 137 hours, 119175 files
• —speech_italian: 70 hours, 59525 files
• —speech_russian: 25 hours, 16566 files
• —speech_spanish: 86 hours, 78603 files
• —noise: 177 hours, 63810 files
• —rir_slr26: 17 hours, 60000 files
• —rir_slr28: 0.1 hours, 248 files
• —License:
• —LibriVox: Public domain
• —AudioSet: CC BY 4.0
• —Freesound: CC0 1.0
• —OpenSLR26 and OpenSLR28: Apache 2.0
• —Source: https://github.com/microsoft/DNS-Challenge/
• —Paper: ICASSP 2023 Deep Noise Suppression Challenge

Usage

import io

import soundfile as sf
from datasets import Features, Value, load_dataset

for item in load_dataset(
    "philgzl/dns5",
    split="speech_english",
    streaming=True,
    features=Features({"audio": Value("binary"), "name": Value("string")}),
):
    print(item["name"])
    buffer = io.BytesIO(item["audio"])
    x, fs = sf.read(buffer)
    # do stuff...

Citation

@article{dubey2024icassp,
  title = {{ICASSP} {2023} {Deep} {Noise} {Suppression} {Challenge}},
  author = {Dubey, Harishchandra and Aazami, Ashkan and Gopal, Vishak and Naderi, Babak and Braun, Sebastian and  Cutler, Ross and Gamper, Hannes and Golestaneh, Mehrsa and Aichner, Robert},
  journal = {IEEE Open J. Signal Process.},
  volume = {5},
  pages = {725--737},
  year = {2024},
}