This project implements a vowel classification system using Support Vector Machines (SVM) in MATLAB. The goal is to build a high-performing classifier for vowel sounds based on acoustic features such as LPC coefficients, duration, and other spectral characteristics.
Unlike typical educational implementations from scratch, this project leverages MATLAB's built-in machine learning tools to focus on maximizing classification accuracy through hyperparameter optimization and careful preprocessing.
For full methodology, results and discussion, see the project report.
The main objective is to accurately classify vowel sounds using SVMs by:
- Extracting informative acoustic features
- Preprocessing and normalizing data
- Tuning hyperparameters extensively
- Evaluating performance with cross-validation
- The model is trained on a custom vowel audio dataset.
- Audio is preprocessed into acoustic features (LPC, duration, etc.)
🔗 Preprocessed Dataset:
Download preprocessed CSV
🎧 Original Dataset (audio files):
Download original data
- ✅ Uses MATLAB's built-in
fitcsvmandfitcecocfor SVM classification - ✅ Experiments with:
- Kernels: linear, RBF, polynomial
- Regularization parameter
C - Kernel scale (
auto, grid search)
- ✅ Feature normalization and scaling
- ✅ Cross-validation for reliable evaluation
- ✅ Final system achieves strong vowel classification accuracy