HMM-Speech-Recognition

This repository contains an implementation of Hidden Markov Models (HMMs) for speech recognition using both discrete and continuous models. The project demonstrates the use of HMMs for sequential data, including synthetic speech data and feature extraction using MFCCs.

---

Features

• Discrete HMM: Implementation of the forward, backward, Viterbi, and Baum-Welch algorithms for discrete observations.
• Gaussian HMM: Continuous observation HMM using multivariate Gaussian distributions for speech features.
• Speech Recognition: Recognizes words based on MFCC features extracted from audio.
• Feature Extraction: Uses librosa to compute MFCCs, delta, and delta-delta features.
• Visualization: Plots transition matrices, observation matrices, initial state distributions, and state transition diagrams.
• Synthetic Data: Demonstrates functionality using synthetic observation sequences and synthetic speech data.
• Open for Extension: Can be extended with real audio datasets, more sophisticated HMM topologies, and integration with language models.

---

Installation

HMM Speech Recognition System

Project Overview

This project implements a Hidden Markov Model (HMM)-based speech recognition system using Gaussian Mixture Models (GMMs) for acoustic modeling. It demonstrates the application of HMMs in recognizing speech patterns, particularly focusing on isolated word recognition.

Project Structure

• HMM.py: Contains the HMM and GaussianHMM classes, which define the structure and operations of the Hidden Markov Models.
• SpeechRecognitionHMM.py: Implements the speech recognition system utilizing Gaussian HMMs.
• main.py: The main script that demonstrates the system's capabilities, including training, testing, and visualization.
• README.md: Provides an overview of the project, setup instructions, and usage guidelines.

Dependencies

Ensure you have the following Python libraries installed:

• numpy: For numerical computations.
• matplotlib: For plotting and visualization.
• scipy: Provides statistical functions, including those for Gaussian HMMs.
• librosa: For audio processing and Mel-Frequency Cepstral Coefficients (MFCC) extraction.

1. Clone the repository:

git clone https://github.com/your-username/HMM-Speech-Recognition.git

You can install these dependencies using pip:

```bash
python - m venv venv
python venv/scipts/activate
pip install numpy matplotlib scipy librosa
pip install -r requirements.txt

cd HMM-Speech-Recognition
python hmm_speech_recognition.py



## Project Structure

- `HMM.py` – Contains HMM and GaussianHMM classes  
- `SpeechRecognitionHMM.py` – Speech recognition system using Gaussian HMM  
- `main.py` – Script for demonstration, training, testing, and plotting  
- `README.md` – Project overview and instructions  

---

## Dependencies

- `numpy` – Numerical computations  
- `matplotlib` – Plotting and visualization  
- `scipy` – Statistical functions for Gaussian HMM  
- `librosa` – Audio processing and MFCC extraction  

---

## Key Algorithms

- **Forward Algorithm:** Computes forward probabilities of sequences  
- **Backward Algorithm:** Computes backward probabilities of sequences  
- **Viterbi Algorithm:** Finds the most likely state sequence  
- **Baum-Welch Algorithm:** Estimates HMM parameters for training  

---

## Demonstration

- `demonstrate_hmm()` – Tests discrete HMM on synthetic data  
- `create_synthetic_speech_data()` – Generates synthetic MFCC features for words  
- `demonstrate_speech_recognition()` – Trains and tests Gaussian HMM for speech recognition  
- `plot_hmm_analysis()` – Visualizes HMM parameters and state transitions  

---

## Future Improvements

- Integrate with real audio datasets for speech recognition  
- Implement more sophisticated HMM topologies (left-to-right, ergodic)  
- Add support for larger vocabularies and language models  
- Optimize training using Baum-Welch with multiple samples per word  
- Include GUI or interactive interface for testing real-time audio