Emoti-Spectro

A research module to test various light to heavy pretrained and made-from-scratch CNN models to predict emotions from spectrogram images.

Table of Contents

• Tech Stack
• Project Structure
• How to use?

Tech Stack used

Stack	Tech
Language
Frameworks
Data Processing
Visualization

Project Structure

Emoti-Spectro/
│
├── metrics/                        # Directory for storing model metrics
├── training_logs/                  # Logs from model training sessions
├── training_plots/                 # Graphs and plots of training history
│
├── 00_prep.py                      # Initial data preparation script
├── 01_convertor.py                 # Audio format conversion utility
├── 02_melspectro.py                # Generates Mel-spectrograms from audio
│
├── 03_lightcnn.py                  # Light custom CNN model implementation
├── 04_deepcnn.py                   # Deeper custom CNN model implementation
├── 05_mobilenetv2.py               # MobileNetV2 transfer learning script
├── 06_mobilenetv2_fine.py          # Fine-tuning MobileNetV2
├── 07_deepcnn_rgb.py               # Deep CNN for RGB spectrograms
├── 08_lightcnn_rgb.py              # Light CNN for RGB spectrograms
├── 09_effnetv2.py                  # EfficientNetV2 transfer learning
├── 10_effnetv2_ft.py               # Fine-tuning EfficientNetV2
├── 11_grucnn.py                    # Hybrid GRU-CNN model
├── 12_cnngru_effnetb0_ft.py        # Complex hybrid architecture
│
├── graphmaker.py                   # Utility to create performance graphs
├── result_calc.py                  # script to calculate and display results
└── requirements.txt                # Project dependencies

How to use?

1. Setup Environment

Ensure you have Python installed. Create a virtual environment and install dependencies.

python -m venv venv
# Windows
venv\Scripts\activate
# Linux/Mac
source venv/bin/activate

pip install -r requirements.txt

2. Data Preparation

Run the preparation scripts in order to process your audio dataset into spectrograms ready for training.

python 00_prep.py
python 01_convertor.py
python 02_melspectro.py

3. Train a Model

Select a model script to start training. For example, to train the Light CNN model:

python 03_lightcnn.py

Or to use a pretrained model like MobileNetV2:

python 05_mobilenetv2.py

4. View Results

After training, check the training_plots/ directory for accuracy/loss graphs, or run the result calculator:

python result_calc.py

Author

Chirag Wattamwar