This repository contains code for an HMM-based automatic speech recognition (ASR) system implemented for a university course assignment.

The ASR system implements acoustic and language modeling components to transcribe audio recordings of spoken English into text. Key aspects include:

• Weighted finite-state transducers (WFSTs) to represent the acoustic and language models
• Viterbi algorithm for decoding
• Beam search and pruning for efficiency
• Lexical tree structure and look-ahead techniques
• Unigram and bigram language models

The report describes experiments using this codebase to explore:

• Varying the number of HMM states per phoneme

• Tuning self-loop transition probabilities

• Adding silence models

• Comparing beam pruning techniques

• Lexical tree structure and look-ahead

• Bigram language modeling

See the report for details and quantitative results on these experiments. The code can be adapted to replicate the experiments.

The report provides academic references for the techniques implemented here.

This project was developed by Ammar Khairi and Mukhtar Mohammed as a course assignment.