Clean Sound with AI - Denoising Audio Using CNN

First I want to Thank Dr.Yevgeny Beiderman and my partner for this project Mr. @baruch456

This project aims to clean sound recordings from noisy environments using a Convolutional Neural Network (CNN) based on the CleanUNet model. The application is built using Python and PyQt5 for the graphical user interface (GUI), and it employs the torchaudio library for audio processing.

Table of Contents

• Features
• Installation
• Usage
• Project Structure
• How It Works
• Dependencies
• Credits
• License

Features

• User-Friendly GUI: Interact with the denoising process through an intuitive graphical interface.

• Audio Denoising: Clean noisy audio files using a trained CNN model.

• Visualization: Plot waveforms, spectrograms, and frequency spectra of the original and denoised audio.

• File Management: Easily select input files and output directories for processing.

• 

Installation

To run this project locally, follow these steps:

1. Clone the repository:

   git clone cd clean-sound-ai

2. Install the required Python packages:

   It's recommended to create a virtual environment first:

   python3 -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

   Then, install the dependencies:

   pip install -r requirements.txt

3. Download the CleanUNet model:

   Place the cleanunet_model_conc.pth file in the root directory of the project. You can download it from https://drive.google.com/file/d/1pdoRDIF38rDHmML0R3wWdCw6KjDN6Gk6/view?usp=sharing .

Usage

1. Run the application:

   python main.py

2. Use the GUI:

   • Input File: Click the "Browse" button to select a .wav file from your computer.
   • Output Directory: Click the "Browse" button to select a directory where the cleaned audio will be saved.
   • Clean Audio: Click the "Denoise" button to start the denoising process.
   • Visualization: The original and cleaned audio will be displayed in the waveform, spectrogram, and frequency spectrum plots.

Project Structure

• main.py: Main script to run the GUI application.
• clean_sound_gui.ui: The UI file designed with Qt Designer.
• CleanUNet/: Directory containing the CleanUNet model implementation.
• requirements.txt: List of required Python packages.
• SpyBot.qss: Custom stylesheet for the PyQt5 GUI.
• README.md: This file.

How It Works

1. Model Loading: The application loads the pre-trained CleanUNet model, which is designed to remove noise from audio signals.
2. Audio Processing: The noisy audio file is loaded, processed by the model, and the cleaned version is saved.
3. Visualization: The original and cleaned audio are visualized in the GUI using waveforms, spectrograms, and frequency spectra.

Dependencies

• Python 3.8+
• PyQt5
• matplotlib
• numpy
• scipy
• librosa
• torchaudio
• torch

Install all dependencies via pip using the requirements.txt file provided.

Credits

This project uses the [CleanUNet model] for audio denoising. The GUI is developed with PyQt5, and the audio processing leverages torchaudio and librosa.