Speaker Recognition Project

Overview

This project is a Streamlit application runnning a keras model trained on our team's voices for speaker recognition . It allows users to sign up, log in, and use various functionalities like recording live audio, uploading audio files, predicting the speaker using a pre-trained model, and retrieving past transcriptions. The application also features a database to store user information and transcription history.

Features

1. User Authentication:

   • Signup and login functionality using bcrypt for password hashing.
   • User data stored in a SQLite database.

2. Speaker Recognition:

   • Record live audio or upload audio files for speaker recognition.
   • Transcriptions generated using Google's Speech Recognition API.
   • Speaker prediction using a pre-trained Keras model.

3. Transcription History:

   • Store and retrieve past transcriptions.
   • Download transcriptions as a text file.

Setup

Prerequisites

• Python 3.6+
• Streamlit
• NumPy
• PyAudio
• Librosa
• TensorFlow
• Scikit-learn
• Soundfile
• SpeechRecognition
• Bcrypt
• SQLite3

Installation

git clone https://github.com/yourusername/speaker-recognition-app.git

cd speaker-recognition-app

python -m venv venv

venv\Scripts\activate

pip install -r requirements.txt

streamlit run app.py

Usage

Sign Up

1. Select the "Sign Up" option.
2. Enter your email and password.
3. Click "Sign Up" to create your account.

Login

1. Select the "Login" option.
2. Enter your email and password.
3. Click "Login" to access the application features.

Speaker Recognition

1. After logging in, choose an option to identify the speaker:
   • Upload Audio File: Upload .wav files for prediction.
   • Record Live Audio: Record live audio using your microphone.
   • Retrieve Past Transcriptions: View and download previous transcriptions.
2. Follow the on-screen instructions to upload or record audio.
3. View the predicted speaker and transcription.
4. Download transcriptions if needed.

File Structure

• app.py: Main Streamlit application file.
• setup_database.py: Script to set up the SQLite database.
• requirements.txt: List of required Python packages.
• model_one.keras: Pre-trained Keras model for speaker prediction.
• label_encoder_one.npy: Label encoder for the model.

Database Schema

Users Table

• id: INTEGER, primary key
• email: VARCHAR(50), unique
• password: VARCHAR(60)
• status: VARCHAR(20)
• created_dt: DATETIME, default current timestamp

History Table

• id: INTEGER, primary key
• user_id: INTEGER, foreign key references users(id)
• name: TEXT
• transcription_file: BLOB
• created_dt: DATETIME, default current timestamp
• updated_dt: DATETIME, default current timestamp

Notes

• Ensure the model_one.keras and label_encoder_one.npy files are in the project directory.
• The app uses Google's Speech Recognition API, which requires an internet connection.
• Audio recording functionality requires a working microphone.

Database Setup

To manually set up the SQLite database for user management and history tracking, follow these steps:

Working Directory: db_helpers

1. Make Database: Use make_db.py to create the SQLite database with the required tables (users and history).

   python make_db.py

2. Check Database (Optional): Use check_db.py to check the existing tables and records in the database.

   python check_db.py

Note: These scripts assume the database file (voice_db.db) is created in the same directory as the scripts. Adjust the database file path if necessary.

For training on different/new data, check training.md