Hearoo 🎧 is an intelligent audio analysis platform that blends
cutting-edge deep learning with an interactive user
experience.
Built with TypeScript on the frontend and
Python on the backend, Hearoo leverages a
ResNet-based CNN model trained on the
ESC-50 dataset via Modal.com to classify and
visualize environmental sounds with remarkable performance,
achieving 81.25% accuracy across 50 sound
categories.
Users can upload WAV files to explore detailed
feature visualizations, turning raw sound into
insightful, dynamic visuals.
Designed with scalability, clarity, and creativity in mind,
Hearoo transforms the way we perceive and interpret sound
through AI-powered intelligence.
Core Features
Deep audio classification using CNN
ResNet-based architecture
Mel spectrogram conversion
Real-time enviroment sound detection
Waveform and spectrogram visualizations
Interactive dashboard built with Next.js
Serverless GPU inference hosted on Modal
TensorBoard integration for model - performance
tracking
Technical Highlights
Backend powered by FastAPI and Python
Frontend written entirely in TypeScript
Optimized training using AdamW
Data augmentation with Mixup technique
Batch Normalization for faster convergence
Pydantic validation for robust API requests
TensorBoard integration for model tracking
Modern UI built with Tailwind CSS and Shadcn UI
