Wav2Vec 2.0 based violence scene detection using only audio data

This repository implements Wav2Vec 2.0 based violence scene detection using only audio data.

( http://ieiespc.org/ieiespc/ArticleDetail/RD_R/412815)

Proposed System

The proposed system is as follows. The MediaEval2015 audio signal is input to the pre-trained Wav2Vec 2.0. After that, the violent scenes are detected with a classifier.

The pre-trained Wav2Vec 2.0 used in the experiment can be downloaded from https://dl.fbaipublicfiles.com/fairseq/wav2vec/wav2vec_small.pt

Dataset used in the experiment

MediaEval2015 was used for the experiment.

Experiment

python main.py wav2vec_small.pt media2015(your dataset path) save(your save path)

Reference

1. Wav2Keyword
2. fairseq_Wav2Vec 2.0