🔊 Spiking Neural Networks for Environmental Sound Classification

This project explores Spiking Neural Networks (SNNs) for environmental sound classification using the ESC-10 dataset.
Multiple spike encoding strategies and SNN architectures are implemented and benchmarked with PyTorch + snnTorch.

---

📁 Project Structure

• preprocessing.py 🎧
  Audio loading and preprocessing: log-Mel spectrograms, normalization, padding and data augmentation.

• spike_encoding.py ⚡
  Dataset classes implementing different spike encodings:

  • Delta Modulation
  • Threshold Adaptive Encoding (TAE)
  • Step Forward Encoding

• models.py 🧠
  Spiking neural architectures:

  • Fully-connected SNN (SNNClassifier)
  • Convolutional SNN (C_SNN)

• train_eval.py 📊
  Training loop and evaluation utilities (accuracy, F1-score, confusion matrices).

• env-soundnet.ipynb 🚀
  End-to-end notebook to run preprocessing, training and evaluation
  (main entry point).

---

🧩 Requirements

Install the required dependencies

🎼 Dataset

The project uses the ESC-10 dataset, organized by class as follows. Each folder contains the corresponding .wav audio files.

▶️ How to Run

1. Open env-soundnet.ipynb (recommended: Google Colab ☁️)
2. Configure the dataset path
3. Run the notebook cells sequentially to:
   • load and preprocess audio data 🎧
   • apply spike encoding strategies ⚡
   • train SNN and Conv-SNN models 🧠
   • evaluate performance and analyze results 📊

Trained models are automatically saved to disk 💾.

---

✨ Notes

• The modular design allows easy experimentation with different spike encodings and network architectures.
• Spike trains and neural activity can be visualized using the provided analysis utilities 📈.