HumorMe - An Audio Humor Detection Demo

A full-fledged web application that records audio and classifies whether it's funny or not using a fine-tuned HuggingFace model. The app features real-time audio recording, file upload capabilities, and a modern, responsive UI.

Features

• Real-time Audio Recording: Record audio directly in the browser
• File Upload Support: Upload audio files (.wav, .mp3, .m4a, etc.)
• AI Classification: Uses fine-tuned wav2vec model for humor detection
• Modern UI: Beautiful, responsive design with real-time feedback
• Confidence Scores: Shows detailed confidence scores for predictions
• Audio Validation: Ensures minimum/maximum audio length requirements

Model Performance

The app uses a fine-tuned HuggingFace model (rishiA/humor_model_v4) that achieves 86% accuracy on humor detection tasks.

Installation & Setup

Prerequisites

• Python 3.8 or higher
• pip (Python package installer)
• Modern web browser with microphone access

Quick Start

1. Clone the repository

   git clone <your-repo-url>
   cd humorMe

2. Install dependencies

   pip install -r requirements.txt

3. Run the application

   python app.py

4. Open your browser Navigate to http://localhost:5000

Production Deployment

For production deployment, you can use Gunicorn:

gunicorn -w 4 -b 0.0.0.0:5000 app:app

How to Use

Recording Audio

1. Click "🎤 Start Recording"
2. Speak something funny or serious
3. Click "⏹️ Stop Recording" when done
4. Wait for the AI analysis

Uploading Files

1. Click "📁 Choose Audio File"
2. Select an audio file from your device
3. Wait for the AI analysis

Understanding Results

• 😂 FUNNY: The AI detected humor in your audio
• 😐 NOT FUNNY: The AI detected serious/non-humorous content
• Confidence Score: How certain the AI is about its prediction
• Detailed Scores: Breakdown of funny vs not-funny probabilities

🔧 Technical Details

Backend (Flask)

• Framework: Flask with CORS support
• Audio Processing: librosa for audio preprocessing
• Model Integration: HuggingFace Transformers pipeline
• File Handling: Temporary file management for audio processing

Frontend (HTML/CSS/JavaScript)

• Audio Recording: Web Audio API with MediaRecorder
• File Upload: Drag-and-drop file input
• Responsive Design: Mobile-friendly interface
• Real-time Feedback: Status updates and progress indicators

Audio Processing Pipeline

1. Input Validation: Check file format and duration
2. Preprocessing: Resample to 16kHz, normalize length
3. Model Inference: Pass through fine-tuned wav2vec model
4. Post-processing: Extract confidence scores and predictions

📊 Model Architecture

The app uses a fine-tuned version of Facebook's wav2vec model:

• Base Model: facebook/hubert-large-ls960-ft
• Task: Binary classification (funny vs not funny)
• Training: Custom dataset with class weighting
• Performance: 86% accuracy on test set

🎨 Customization

Styling

The UI uses CSS custom properties and can be easily customized by modifying the styles in templates/index.html.

Model Configuration

To use a different model, update the model name in app.py:

classifier = pipeline("audio-classification", model="your-model-name")

Audio Constraints

Modify audio length constraints in app.py:

# Minimum length (0.5 seconds)
if len(audio_data) < 8000:
    
# Maximum length (30 seconds)
if len(audio_data) > 480000:

Performance Optimization

• Model Caching: Model is loaded once at startup
• Temporary Files: Automatic cleanup of processed audio files
• Audio Compression: Optimized audio preprocessing pipeline
• Async Processing: Non-blocking audio classification

License

This project is open source and available under the MIT License.