EmotionAI Voice is an open-source deep learning project that classifies vocal emotions using raw .wav audio.
It’s designed for applications in mental health monitoring, UX analysis, and intelligent speech interfaces.

🔬 The model is trained from scratch, using spectrogram-based audio features, and aims to recognize 8 core emotions.

🎯 Features

• 🧠 Emotion recognition: neutral, calm, happy, sad, angry, fearful, disgust, surprised
• 🎧 Accepts .wav audio inputs (from RAVDESS dataset)
• 📊 CNN and CNN+GRU models implemented in PyTorch
• 🔍 Real-time evaluation with confusion matrix and accuracy tracking
• 🛠️ Fully open-source and customizable (no pre-trained models)
• 🧪 Includes SpecAugment for data augmentation (frequency/time masking)

📚 Dataset — RAVDESS

We use the RAVDESS dataset, which includes:

• 🎭 24 professional actors (balanced male/female)
• 🎙️ 1440 .wav files (16-bit, 48kHz)
• 8 labeled emotions:
  neutral, calm, happy, sad, angry, fearful, disgust, surprised

Each .wav file is preprocessed into a Mel spectrogram and stored as .npy format.

🧠 Model Architectures

2 different models

✅ CNN (Best Performance)

• 3x Conv1D + ReLU + MaxPool
• Fully connected layers
• Dropout regularization (adjustable)

🔁 CNN + GRU

• CNN front-end for spatial encoding
• GRU (recurrent layers) to capture temporal dynamics
• Lower accuracy than CNN-only model

🧪 SpecAugment: Data Augmentation

To improve generalization, we implemented SpecAugmentTransform which applies:

• 🕒 Time masking: hides random time intervals
• 📡 Frequency masking: hides random mel frequency bands

📈 Training Results

• Best Validation Accuracy: ~49.6%
• Training set: Actors 1–20
• Validation set: Actors 21–24

Confusion Matrix Example:

🔍 Key Observations:

• Surprised, calm, and disgust are the most accurately predicted emotions.
• Neutral, happy, and sad tend to be confused with each other, which is common due to subtle acoustic variations.
• The model struggles with fearful and angry in some cases — suggesting those may share overlapping vocal characteristics in this dataset.
• Emotion classes like happy and fearful are often misclassified due to variability in expression intensity among different actors.

📈 Interpretation

While the model captures general emotion cues, it suffers from class overlap and limited generalization. The accuracy remains significantly above random (12.5% for 8 classes), but there is still room for improvement.

🚀 Getting Started

1. Install dependencies

pip install -r requirements.txt

2. Download dataset from Kaggle

Follow the instructions in the README.md located in the data folder

3 . Train the model

python src/train.py

4. Evaluation the performances with a confusion matrix

```bash python src/confusion_matrix.py

A smart computer vision system that detects signs of fatigue, eye strain, and microsleep using a standard webcam.

🎯 Purpose

FatiguEye is a real-time fatigue detection system based on eye tracking and facial landmark analysis.
It helps identify early signs of drowsiness by measuring:

• 👁️ Eye Aspect Ratio (EAR)
• 🔁 Blink frequency
• ⏱️ Prolonged eyelid closure
• ⚠️ Microsleep events

Ideal for driver monitoring, industrial safety, or ergonomic fatigue prevention.

🧠 How It Works

FatiguEye uses MediaPipe Face Mesh to extract eye landmarks, and computes the EAR (Eye Aspect Ratio) on each video frame.

📡 Processing pipeline:

1. 🎥 Webcam feed is captured in real-time
2. 🧠 Facial landmarks (eyes) are detected with Mediapipe
3. 📏 EAR is calculated per eye
4. 🧮 Blink count and eye closure duration are analyzed
5. 🔔 Fatigue alerts are raised (visual + audio)

🚀 Demo Preview

💻 Technologies Used

📦 Installation

```bash git clone https://github.com/Tirovo/fatigueye.git cd fatigueye python -m venv venv source venv/bin/activate # Or venv\Scriptsctivate on Windows

BoardMapper is an open-source tool designed to automatically generate PCB layout. It labels component references (e.g. U1, R1, C1) directly on the circuit image, facilitating component identification for reverse engineering purposes.

🎯 Purpose

• 🤖 Automation: Eliminates the need for manual placement annotation on PCB layouts.
• ⏱️ Efficiency: Saves time for engineers and makers working on PCB assembly and debugging.
• 🔍 Clarity: Provides a clear visual reference for debugging, testing, and manufacturing.
• 💻 Cross-Platform: Works on Windows, Linux, and macOS systems.

📝 Annotation

Position	Original	Annotated
Top
Bottom

📋 Requirements

• Python: Version 3.6 or higher
• Required Libraries:
  • opencv-python (for image processing)
  • lxml (for XML parsing)

🚀 Installation Instructions

🛠️ Setup

1. Clone the repository or Download the project to your local machine.

2. Labeling the PCB:
   • Step 1: Take a photo of both the top and bottom layers of the chosen PCB.
   • Step 2: Place the top.png and bottom.png images into the input folder.
   • Step 3: Install the latest version of LabelImg.
   • Step 4: Open top.png in LabelImg and draw bounding boxes around each component. Label each component according to its type:
     • R: Resistor
     • C: Capacitor
     • L: Inductor
     • F: Fuse
     • POT: Potentiometer
     • D: Diode
     • LED: LED
     • Q: Transistor (BJT, MOSFET)
     • U: Integrated Circuit (IC)
     • J: Connector
     • K: Relay
     • SW: Switch
     • Y: Quartz / Resonator
     • SP: Speaker
     • ANT: Antenna

     LabelImg Shortcuts:

     • ✏️ W: Draw a new rectangular bounding box (RectBox)
     • ❌ D: Delete the last drawn bounding box
     • 💾 Ctrl + S: Save the annotation as an XML file
     • ⏪ Ctrl + Z: Undo the last action
     • 📋 Ctrl + C: Copy a bounding box
     • 📏 Ctrl + V: Paste a copied bounding box
     • 🔍 Ctrl + A: Select all bounding boxes
     • 🔄 Ctrl + R: Rotate the image (for better labeling)
     • 🚫 Esc: Cancel the current operation or close a dialog box
   • Step 5: After labeling the top.png, save the annotation as top.xml.
   • Step 6: Repeat the labeling process for the bottom.png and save it as bottom.xml.
   • Step 7: Place both top.xml and bottom.xml into the input folder.
3. Running the Tool:
   • Windows: Double-click on setup_and_run.bat to automatically run the script. The tool will read the XML annotations, draw bounding boxes on the images, and save the annotated images.
   • Linux/macOS: You can run the script from the terminal:

     chmod +x script.sh
     ./script.sh
     

4. 📂 Output:
   • After the script has executed, navigate to the output folder to find the resulting annotated images:
     • top_annotated.png
     • bottom_annotated.png

🤝 Contributions

If you’d like to contribute to the project, please follow these steps:

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Push to the branch
5. Open a pull request

We welcome any contributions to improve BoardMapper! 🎉

VATN (from the Old Norse word for “water”) is a game designed to raise awareness about water management. Players must make critical daily decisions to address major water-related issues in their country, influencing key parameters that determine the nation’s survival.

🎯 Purpose

• 💧 Water Management Awareness: Educate players on the importance of sustainable water policies.
• 🎮 Engaging Decision-Making: Every day presents a new challenge that affects the nation’s status.
• 🏆 Survival & Strategy: Keep your country alive as long as possible by maintaining stability.
• 📊 Progress Tracking: Visualize the evolution of key parameters through in-game graphs.

📝 Features

📺 Gameplay Overview

🎮 Main Menu	📊 In-Game Statistics	🏆 Endgame Rankings

Anti-Vuvuzela Filter is an open-source project dedicated to second-order analog filters and beyond. This project was initially developed to design an “anti-vuvuzela” filter, aiming to attenuate the distinctive and persistent sound of vuvuzelas while preserving the clarity of commentators’ voices during the 2010 FIFA World Cup.

🎯 Purpose

• 🔇 Targeted Noise Reduction: Specifically designed to attenuate vuvuzela noise while maintaining the intelligibility of speech.
• 🎚 Second-Order Analog Filtering: Utilizing advanced filtering techniques for efficient noise cancellation.
• 🛠️ Open-source and Customizable: Modify and adapt the design for other audio filtering applications.

📝 Features

📐 Simulation & Testing

🛠️ LTSpice Circuit	📜 Simulation

The Motor Control Shield is an open-source project designed for controlling DC motors. It comes in the form of an Arduino shield mounted on an STM32 Nucleo board. The shield enables motor control via an NMOS transistor, current measurement with a shunt resistor, and rotation tracking using data from an incremental encoder.

🎯 Purpose

• 🔄 Motor Control: Provides precise control over DC motors.
• 📉 Current Measurement: Monitors the current consumed by the motor.
• 🔄 Rotation Tracking: Uses an incremental encoder to track motor rotation.
• 🛠️ Open-source & Customizable: Modifiable and adaptable for various projects.

📝 Features

📐 PCB Design Preview

📜 Functional diagram	📜 Schematic	🖥️ PCB Layout	🏗️ 3D