7tl7qns7ch / SNU-B36-EX

Zero-Shot Single-Microphone Sound Classification and Localization in a Building via the Synthesis of Unseen Features (IEEE Transactions on Multimedia 2021).

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

SNU-B36-EX

SNU-B36-EX is an inter-floor noise dataset which is extended version of SNU-B36-50 collected in building no.36 at Seoul National University, Seoul, Korea. It was recorded with a smartphone (Samgsung Galaxy S6) with sampling rate 44,100Hz. The data set is available at SNU-B36-EX. The name of each folder indicate class number (see Train/test split for Zero-Shot settings , the bottom of this page).

All setting and procedure are same as the previous works, but it has the additional noise data by setting the horizontal source points in 1m increments from 0 to 12m. Thus, there are 8,450 audio files in the data set.

The dataset consists of 5 types:

'MB' : a medicine ball on the floor,

'HD' : dropping a hammer on the floor,

'HH' : hitting with a hammer on the floor,

'CD' : dragging a chair on the floor,

'VC' : operating a vacuum cleaner,

and 39 positions:

3F0m, 3F1m, 3F2m, 3F3m, 3F4m, 3F5m, 3F6m, 3F7m, 3F8m, 3F9m, 3F10m, 3F11m, 3F12m,

2F0m, 2F1m, 2F2m, 2F3m, 2F4m, 2F5m, 2F6m, 2F7m, 2F8m, 2F9m, 2F10m, 2F11m, 2F12m,

1F0m, 1F1m, 1F2m, 1F3m, 1F4m, 1F5m, 1F6m, 1F7m, 1F8m, 1F9m, 1F10m, 1F11m, 1F12m.

Train/test set split for Zero-Shot settings

Train set is colored by black, test set is colored by red, and numbers colored by green are empty set.

Citation

@article{lee2021zssl,
  title={Zero-Shot Single-Microphone Sound Classification and Localization in a Building Via the Synthesis of Unseen Features},
  author={Seungjun Lee, Haesang Yang, Hwiyong Choi, and Woojae Song},
  journal={IEEE Transactions on Multimedia},
  year={2021}
}

About

Zero-Shot Single-Microphone Sound Classification and Localization in a Building via the Synthesis of Unseen Features (IEEE Transactions on Multimedia 2021).

License:MIT License


Languages

Language:Python 92.7%Language:MATLAB 6.5%Language:Shell 0.7%