Non-Speech Sound Classification

This project focuses on classifying non-speech sounds into seven distinct categories using a Convolutional Neural Network (CNN). The dataset used consists of 7,000 audio samples, and class imbalance was addressed as part of the preprocessing pipeline.

Model Highlights:

• Implemented a custom CNN architecture for audio classification
• Applied techniques to correct class imbalance
• Evaluated on a held-out test set with the following performance:

Metric	Value
Test Loss	0.5746
Test Accuracy	82.48%
Precision	83.91%
Recall	82.48%
F1 Score	82.69%

Results:

Tools and Libraries:

• PyTorch
• SciPy

References:

• 7K Non-Speech Sound Dataset Paper
• Challenge Link
• Dataset on Zenodo

---

Let me know if you'd like a LaTeX version of this for a resume, report, or poster.