jakobabesser / embedding_robustness_2022

How Robust are Audio Embeddings for Polyphonic Sound Event Tagging?

Reference

This jupyter notebook reproduces all experiments and figures for the article:

Jakob Abeßer, Sascha Grollmisch, Meinard Müller, How Robust are Audio Embeddings for Polyphonic Sound Event Tagging? (2022)

How to reproduce experiments / figures?

• download supplementary data from https://zenodo.org/record/7912746 and store files in the data subfolder, here's the list of required files:

  • all_emb.p - md5:88b01701429904b8f5aa2b6338f51f54 - 2.2 GB
  • class_id.npy - md5:960d8f50da888ef4e7b8b003c505a544 - 16.1 kB
  • class_labels.p - md5:4e686fd87b8d96a87400598bdf98e8f7 - 810 Bytes
  • prefix_list.p - md5:f3e7f9f04ac220950d74f53847951c87 - 926.0 kB

• install Miniconda on your system

• create a new conda environment including the following packages with the defined versions

python version: 3.7.4
jupyter version: 1.0.0
numpy version:  1.18.5
librosa version:  0.8.1
pandas version:  1.3.1
matplotlib version:  3.4.3
tensorflow version:  2.3.0

• activate conda environment

• run jupyter notebook

• open article_2023_reproduction.ipynb and run all cells

• result figures will be stored in results folder

Details on extracted embeddings

• Details on extracted embeddings

ESC50Mix dataset

• The dataset has been published at https://zenodo.org/record/7913031