Project information
- Category: AI/ML
- Project date: 2022
- Project URL: github.com/jrtivald/EEC201-Final-Project
Voice Classifier
This project used a form of supervised learning to train a classifier and recognize different speakers' voices. Linde Buzo Gray (LBG) Vector Quantization (VQ) was the algorithm used for classifying voices and Mel-frequency cepstral coefficients were the unique features extracted from a speaker's voice. We used tagged data to train our classifier by finding the centroid of a speaker's voice in the multi-dimensional feature space. The feature space centroids were essentially a fingerprint for each speaker's voice. Then as unknown audio is played through the algorithm, calculated centroids are compared to the 13 speaker's centroid fingerprints to determine the likelihood of each speaker generating the new audio file. The uncertainty matrix above shows that we were able to positively classify each speaker after the centroid fingerprint had been trained.