Teleprompter

Browser-based teleprompter and pronunciation trainer for public speakers. Amber phosphor broadcast aesthetic.

Stack

• Frontend — React 18 + Vite, inline CSS, Vitest unit tests
• Backend — FastAPI + edge-tts (TTS) + faster-whisper (speech recognition), served via uvicorn
• Packaging — Multi-stage Docker image (node build → python runtime)

Panels

Speaker

Teleprompter with smooth auto-scroll and TTS narration.

Action	Keyboard	UI
Play / Pause scroll	Space	PLAY button
Speak / Stop TTS	T	SPEAK button
Speed (scroll & TTS)	↑ / ↓ (±0.1x)	Slider (0.1x – 3.0x)
Font size	[ / ]	A− / A+ buttons
Reset	R	RESET button
Mirror	M	MIRROR button

Speed adjusts live during TTS playback — voice and scroll respond immediately. Select text before pressing T to speak only the selection. Text can be loaded from a file or pasted directly.

Tutor

Pronunciation practice: load reference text, speak into the mic, compare your speech to the original with color-coded word-level diff.

• Green — correctly recognized words
• Amber — extra words (added/substituted)
• Red strikethrough — missing words (skipped)

Includes a chronometer that starts on first word and stops when finished. Whisper model and language are selectable. Mic and audio output can be switched live.

Free Speech

Open-ended transcription without a reference text. Accumulates a running transcript across multiple recording sessions. Includes a chronometer and live audio level meter. Useful for rehearsing unscripted speeches.

Run locally

# Backend (from backend/)
uv run uvicorn main:app --reload   # http://localhost:8000

# Frontend (from frontend/)
npm install
npm run dev                        # http://localhost:5173 (proxies /api to :8000)
npm test                           # unit tests

See backend/README.md for API reference, audio pipeline details, and concurrency model.

Run with Docker

docker build -t teleprompter .
docker run -p 8000:8000 teleprompter   # http://localhost:8000

Speech file format

## Section Title     → amber section header
**Bold line**        → emphasized text
---                  → visual spacer
Regular line         → normal paragraph

Files are loaded via the browser FileReader API — never uploaded to the server.