The idea
Everyone’s voice is unique. Speakify identifies a speaker from a short audio clip — useful for things like voice-based biometric identification.
What I built
A deep-learning pipeline that turns audio into MFCC features and spectrograms (visual representations of a sound’s frequencies), then classifies the speaker with a convolutional neural network combined with Gaussian Mixture Models.
To make the model usable, I built an application around it with a Flask backend and a React front end, so it can be applied to different scenarios such as identity verification.
Outcome
A complete research-to-product build — from feature extraction and model design to a working app — completed as my final-year project at FAST.