Voice-Authentication-System

1. Project Description

This project implements a basic voice-based authentication system leveraging machine learning techniques. It uses audio feature extraction (MFCC and its derivatives) to model and recognize a specific speaker’s voice. The system enables a user to enroll with their voice and then authenticate future voice inputs using a trained Support Vector Machine (SVM) classifier.

The application is ideal for exploring the fundamentals of speaker recognition and machine learning-based audio processing.

2. Tech Stack Used

Programming Language: Python

Libraries:

1. librosa – for audio loading and MFCC feature extraction

2. scikit-learn – for feature scaling and SVM classification

3. joblib – for saving and loading models

4. numpy – for numerical operations

5. Google Colab – for easy file uploads and execution

3. Features

• User Enrollment: Uploads two .wav files (user and non-user), extracts features, and trains an SVM model.

• Voice Authentication: Authenticates a new voice sample against the trained model and returns success or failure with confidence.

• Feature Extraction: Uses MFCC along with delta and delta-delta features to enhance speaker recognition.

• Persistence: Saves both the trained model and scaler to disk for future use.

4. How It Works

1. User Enrollment:

• The user uploads:

  • Their own .wav voice sample

  • A non-user .wav sample

• MFCC + delta features are extracted from both.

• A binary SVM classifier is trained and saved.

2. Authentication:

• The user uploads a new .wav sample.

• The model extracts features and predicts whether it matches the enrolled user.

• Authentication success/failure is reported based on probability threshold.