EYEM — Eye-Blink Morse Translator

Turn eye blinks into Morse code and text — live, in your browser.
Private by design. No video ever leaves your device.

EYEM turns eye blinks into Morse and text instantly in your browser, private and serverless. Auto-calibrates, supports 54 symbols, live two-way translation, Morse beep, voice-to-Morse, quick copy, collapsible table and C/M/F/R shortcuts with a dark purple accessible UI for inclusive assistive needs.

---

🌐 Live Demo

Just open the link — no install required:

https://tk-2f.github.io/EYEM/

1. Press Start Camera (or C) → allow camera permission
2. Keep eyes open 2–3s — purple vertical bar calibrates, then disappears
3. Blink to type — translation appears instantly
4. No camera? Type directly or press M to speak

Requires Chrome/Edge, https:// or localhost, and camera/mic permission.

---

📸 Preview

---

✨ Features

• Real-time blink → Morse → Text — adaptive auto-calibration (~2.8s, 45 EAR samples), 5-frame smoothing, hysteresis thresholds
• Bidirectional — type Morse to get English, type English to get Morse instantly
• 54 symbols — A–Z, 0–9 and punctuation . , ? ' ! / ( ) & : ; = + - _ " $ @
• Single-line translation table — Letters / Numbers / Punctuation, each collapsible (▼), every cell in one line like A .-
• Voice → Morse — mic button inside Translation box + M shortcut (Web Speech API)
• Morse audio — dot 80ms / dash 240ms at 700Hz, mute toggle inside Morse box (persists via localStorage, respects prefers-reduced-motion)
• Copy & Sound inside boxes — Copy bottom-left, Voice/Sound bottom-right, with light click pop (900→600Hz)
• Notifications — toasts + Quick Start slide from the right (top:58px), 5s delay, auto-hide 20s, × on hover
• Dark purple accessible UI — var(--muted) #a99ac0, custom 8–10px scrollbar, eye-indicator (top-right), vertical calibration bar (#9f67ff → #5b21b6)
• Keyboard shortcuts — C Camera, M Mic, F delete last symbol, R reset (all ignored while typing)
• 100% client-side — MediaPipe runs in-browser, no backend, no data upload

---

🎯 How It Works

1. Camera on → MediaPipe Face Landmarker (Tasks Vision 0.10.14, GPU→CPU fallback) tracks 478 landmarks locally via CDN
2. EAR calculation → eye-aspect-ratio smoothed over 5 frames; thresholds 0.62× / 0.78× median with hysteresis
3. Timing → <0.15s ignored (natural blink) · 0.15–0.70s = dot · · 0.70–2s = dash — · >2s ignored
4. Sequencing → Double blink (700ms gap) = new letter (space) · Triple blink = new word (/) · Flush idle 750ms
5. Translation → js/morse.js decodes live, updates both boxes and aria-live region

---

⌨️ Blink & Typing Guide

Action	Result
Quick blink (~0.15–0.70s)	· dot
Hold ~1s (0.70–2s)	— dash
Double blink	New letter (space)
Triple blink	New word (/)
F	Delete last symbol (·/—/ / /)
R (camera on, outside input)	Clear all — Cleared — blink again to start.
In Morse box	Type . - · space = letter · // or double-space = word
In Translation box	Type English → Morse appears instantly
C / M	Toggle Camera / Mic (outside inputs only)

Example: .... . .-.. .-.. --- / ..-. .-. .. . -. -.. → HELLO FRIEND

---

🛠️ Tech

• Frontend — Vanilla HTML/CSS/JS, no build step, versioned cache-bust (?v=52)
• Vision — @mediapipe/tasks-vision@0.10.14 + WASM (face_landmarker.task from storage.googleapis.com)
• Audio — Web Audio API (click + Morse beep), Web Speech API (webkitSpeechRecognition fallback)
• Perf & A11y — 5-frame EAR average, GPU→CPU fallback, aria-live="polite" Morse feed, focus trap + Esc in About modal, prefers-reduced-motion guard

Run locally (camera/mic need secure context):

# Python
python -m http.server 8000
# then open http://localhost:8000

# or Node
npx serve .

---

🔒 Privacy

Everything runs on-device. No video, audio, or keystrokes are sent to any server. MediaPipe and Speech Recognition run in your browser.

---

👥 Who Is EYEM For?

• People with limited mobility (ALS, spinal injuries) who can blink reliably
• Assistive-technology builders exploring camera-based input without special hardware
• Developers / students / researchers in CV, HCI, eye-tracking
• Morse learners and anyone curious to control software with eyes

---

🙏 Credits

Developed by Tareq khattab

• GitHub: @tk-2f

---

_{Built with care — accessibility first. If EYEM helped you, please ⭐ the repo.}