CSC 375 Final Project

Established paper: Link to Paper

---

Baseline model

Dependencies

• Python 3.8+
• torch 2.7.0+cu118
• torchaudio 2.7.0+cu118
• torchvision 0.22.0+cu118
• pandas 2.2.3
• numpy 2.2.5
• matplotlib 3.10.1
• scikit-learn 1.6.1
• librosa 0.11.0

How to run

python train_concat_model.py

---

CNN+Transfomer Hybrid model

Dependencies

• Python 3.8+
• torch 2.7.0+cu118
• torchaudio 2.7.0+cu118
• torchvision 0.22.0+cu118
• pandas 2.2.3
• numpy 2.2.5
• matplotlib 3.10.1
• scikit-learn 1.6.1
• tqdm 4.65.0

How to run

python precompute_mfccs.py
python train_transformer.py

---

Improved CNN+LSTM model (not adopted because of time constraints)

This script performs 10-fold cross-validation on the UrbanSound8K dataset, saves the best model for each fold, and generates plots for metrics such as loss, accuracy, macro F1-score, precision, and recall.

Dependencies

• Python 3.8+
• audioread 3.0.1
• librosa 0.11.0
• matplotlib 3.10.1
• numpy 2.2.5
• pandas 2.2.3
• scikit-learn 1.6.1
• scipy 1.15.2
• soundfile 0.13.1
• torch 2.7.0+cu118
• torchaudio 2.7.0+cu118
• torchvision 0.22.0+cu118

How to run

python train_improved_concat_model.py