theopsall/archeo

ARCHEO: a python lib for sound event detection in areas of touristic Interest

archeo

ARCHEO: a python lib for sound event detection in areas of touristic Interest.

Archeo dataset contains small audio files from 2 main data sources. The first source comes from recordings in the urban area of Athens while the second from corresponding videos, of a walk in the urban area of Athens, from YouTube. The data set consists of 13 classes in total as described in the publication.

Link: https://drive.google.com/file/d/1rxLlUlqj72oU2Uz9VWQREcBB3Am6YBHr/view?usp=sharing

---

Experiments

Experiments, as described in the paper.

1 Train Multi Label SVC Classifier based on audio features

    python3 audio_features_classifier.py -a /home/SOURCE_DATA/ -g /home/SOURCE_LABELS/

2 Train Multi Label SVC Classifier with smote based on audio features

    python3 smote_audio_feature_classifier.py -a /home/SOURCE_DATA/ -g /home/SOURCE_LABELS/ -res 2000

3 Train Multi Label SVC Classifier based on Bag of Visual Words descriptors extracted from spectograms

    python3 bag_visual_word_classifier.py -i /home/SPECT/ -o /home/bovw

4 Train Multi Label Convolutional Neural Networks based on Spectograms images

    python3 cnn.py -i /home/SPECT/

---

In order to run the last two experiments, you first have to seperate the wavs files into subdirectories and then create the spectrograms.

Seperate wavs to subdirectoies base on the label

python3 source/split_wavs.py  -a home/DATA_SOURCE -g home/DATA_LABELS -o WAVS

Create spectrograms

python3 source/make_spectros.py  -a home/WAVS -o SPECT