Succor

Your AI backup when every second counts.

Comment

Succor
Your AI backup when every second counts.

📄 Project Description

🚨 Problem I'm Solving

Every year, emergency calls are mishandled due to dispatcher errors caused by stress, human limitations, or communication issues. These delays can cost lives. I wanted to explore how AI could help support dispatchers in real-time.

👤 Who It’s For

  • Trainee 911 dispatchers
  • Emergency response educators
  • Developers interested in voice-based AI
  • Anyone who wants to practice emergency simulations using just their voice

⚙️ How It Works

  1. I speak into the mic to simulate a 911 call.
  2. Succor transcribes my speech using Speech-to-Text.
  3. It analyzes my voice for signs of emotion (nervousness, panic, calm).
  4. It detects keywords, classifies the type of emergency, and estimates the likelihood of it being real, hidden, or a prank.
  5. It visually highlights potential injuries and suggests what to do next — all in real-time.

❤️ Why It Matters

This project helped me appreciate how demanding a dispatcher’s job is. By combining voice emotion and language processing in the browser, Succor could eventually serve as a tool to train dispatchers, reduce mistakes, and save lives.

🚀 What It Does

  • 🎙️ Listens to voice input like a real 911 call
  • ✍️ Transcribes speech to text live
  • 📈 Analyzes emotional tone using audio features
  • 🧠 Detects and classifies emergency scenarios
  • 🩻 Highlights potential injury zones
  • 📊 Adjusts probabilities of the situation type
  • 💻 Runs entirely in the browser

⚙️ Tech Stack Used

  • Languages: JavaScript, JSX
  • Frameworks: React (Next.js), Tailwind CSS,
  • Libraries & APIs:
    • Web Speech API (for STT)
    • Meyda.js for voice emotion analysis
    • Custom JSON logic for accident classification

🛠️ How I Built It

I used Next.js for the frontend and designed everything with Tailwind CSS and Framer Motion for a dark, sci-fi theme. I integrated the Web Speech API for continuous speech recognition and Meyda.js for voice signal analysis. I created a JSON-based classification system that reacts to transcribed text and inferred emotions, updating visuals and probabilities live.

⚔️ Challenges I Ran Into

  • Getting STT to run continuously
  • Syncing audio feature extraction and live transcription
  • Designing meaningful emotion-to-risk logic
  • Making everything work without a backend or external ML engine

🏆 Accomplishments I’m Proud Of

  • Built a fully working voice-based emergency assistant
  • Integrated emotion detection and speech analysis together
  • Created a live simulation tool that's both useful and educational
  • Designed it to be browser-only — no installs or servers needed

📚 What I Learned

  • How to analyze voice features like pitch and energy using JavaScript
  • How dispatchers think under pressure
  • How to structure real-time UI updates in React
  • The power of combining audio and language data in one interface

🔮 What’s Next for Succor

  • 🧠 Add machine learning with TensorFlow.js for better emotion detection
  • 🌐 Multilingual support
  • 🔊 Text-to-Speech (TTS) responses
  • 🧪 Training mode with simulated calls
  • 📱 Mobile-first redesign so anyone can train anywhere

Built With

Share this project:

Updates

Submission history