VGGish

VGGish are features from a pretrained CNN by Google (research paper). Apple has a nice comprehensible explanation.

They benchmark their approach against Audio Set (obs innehåller även djurljud!). It seems to be just tags to YouTube videos?

• AED: Acoustic Event Detection
• VGGish: Seem to be feature from a pretrained CNN? Not sure, but link to repo here

Where do I find the code?

Currently working on data preprocessing in data.ipynb.

How to run

1. Download the OpenMIC-2018 dataset and add as a subfolder to data (data/openmic-2018/all/goes/here)
2. Make sure you have Docker up and running, and open the project in Visual Studio Code.
3. VSC should prompt for Open project in .devcontainer?
4. Accept.