Afterglow

Turn any video into an interactive shadowing workstation.

AI-powered smart segmentation · Single-sentence looping · Hands-free hotkeys · 100% local & offline

English | 中文

---

What is Afterglow?

Afterglow is a local-first language learning player designed for shadowing practice. Drop in any video file, and Afterglow will:

1. Transcribe it into timestamped sentences using Faster-Whisper
2. Segment the speech into logical sentence boundaries
3. Play each sentence with configurable looping and gap pauses — so you can listen, pause, and repeat

No cloud APIs. No subscriptions. Everything runs on your machine.

---

Demo

---

Features

Feature	Description
AI Transcription	Powered by Faster-Whisper (base model, ~143 MB). Auto-downloads on first run.
Single-Sentence Loop	Repeat current sentence 1 / 2 / 3 / 5 / ∞ times before advancing.
Smart Gap	Auto-pause between sentences — giving you time to shadow.
Hotkeys	Space play/pause · Enter replay · ←→ prev/next · ↑↓ speed
Segment Merge	Combine broken segments with one click.
Subtitle Blocker	Hide embedded subtitles so you rely on your ears.
Smart Caching	SHA-256 file hash → skip re-transcription on reload.
Playback Speed	0.3× to 3.0× in 0.1× steps.

Playback Mode Matrix

	Gap OFF	Gap ON
Loop OFF	Normal continuous play	Auto-pause after each sentence
Loop ON	Repeat N times, then next	Repeat → pause → repeat cycle

---

Quick Start

Prerequisites

• Python 3.13+
• Node.js 22+ (with npm)

One-Click Launch (Windows)

git clone https://github.com/your-username/Afterglow.git
cd Afterglow
start.bat

The script installs dependencies, finds free ports, starts both services, and opens your browser.

Manual Launch (Any OS)

Backend:

cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Frontend:

cd frontend
npm install
npm run dev -- --port 5173

Then open http://localhost:5173 in your browser.

---

Project Structure

Afterglow/
├── frontend/                # React 19 + TypeScript + Vite
│   ├── src/
│   │   ├── components/      # UI components (VideoPlayer, TranscriptPanel, ...)
│   │   ├── hooks/           # Custom hooks (useShadowPlayer, useHotkeys, ...)
│   │   ├── services/        # API client & file hashing
│   │   └── types/           # TypeScript interfaces
│   └── vite.config.ts
│
├── backend/                 # FastAPI + Faster-Whisper
│   ├── app/
│   │   ├── routers/         # /api/transcribe, /api/cache
│   │   └── services/        # Whisper model wrapper
│   └── requirements.txt
│
├── docs/                    # Requirements & plans
├── start.bat                # Windows launcher
└── README.md

Runtime data is stored in ~/.afterglow/:

~/.afterglow/
├── cache/       # Transcription cache (JSON, keyed by SHA-256)
└── models/      # Whisper model files (auto-downloaded)

---

API Reference

Method	Endpoint	Description
POST	/api/transcribe	Upload video file → get timestamped segments
GET	/api/cache/{hash}	Retrieve cached transcription by file hash
PUT	/api/cache/{hash}	Store transcription result in cache

---

Tech Stack

Layer	Technology
Frontend	React 19, TypeScript, Vite 7, CSS3
Backend	Python 3.13, FastAPI, Uvicorn
ASR Engine	Faster-Whisper (CTranslate2)
Caching	SHA-256 hash → local JSON files

---

Roadmap

• Core video player with transcript panel
• AI transcription (Faster-Whisper)
• Single-sentence loop & smart gap
• Hotkey controls
• Segment merging
• Transcription caching
• Recording & playback comparison
• Pronunciation scoring
• Desktop app packaging (Electron / Tauri)

---

License

This project is licensed under the CC License.

---

_{Built for language learners who believe in deliberate practice.}