Diploma-thesis

Title: "Investigation of acoustic scene classification techniques"

Versions used:

Run Training_stage.ipynb to load filenames and labels, pre-process your data (choose either Per-Channel Energy Normalization or log-Mel power spectrograms) and train your model. Multiple CNN architectures are available in the repository (see corresponding Python scripts).
Run MLP+ETi.ipynb for feature extraction and Enhanced Temporal Integration. An efficient MLP architecture is provided for training (can be trained on CPU, ~ 1 sec per epoch).
Run grad-CAM.ipynb for 'Gradient-weighted Class Activation Mapping' visualization technique. Make sure you use the same pre-processing step throughout this process.

Also, keep an eye on the comments since they provide useful explanations for every step.

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
Attention_ResNet.py		Attention_ResNet.py
FCN _1.py		FCN _1.py
FCN_2.py		FCN_2.py
LICENSE		LICENSE
MLP+ETi.ipynb		MLP+ETi.ipynb
Mixup_class.py		Mixup_class.py
README.md		README.md
ResNet.py		ResNet.py
ResNet_split_frequencies.py		ResNet_split_frequencies.py
Training_stage.ipynb		Training_stage.ipynb
grad-CAM.ipynb		grad-CAM.ipynb