Recognizing Spoken Arabig Digits

Mission

The objective of this project was to utilize Gaussian Mixture Models (GMMs) to create a model capable of identifying spoken Arabic digits (0-9) based on their unique phonetic makeup. Each digit is characterized by a distinct set of phonemes, represented in the spoken Arabic digit dataset by unique sets of cepstral coefficients.

To accomplish this task, I successfully implemented Markov clustering with a GMM generated using k-means clustering and expectation-maximization algorithms. Subsequently, I conducted experiments with various probabilistic models, aimed at investigating and comprehending the impact of different modeling choices on the digit identification task.

​

​