mirror of
https://github.com/Memo-2023/mana-monorepo.git
synced 2026-05-22 16:06:43 +02:00
4b08c41547
Integrate new transcriber application for AI-powered YouTube video
transcription with full monorepo structure and Groq Whisper API support.
## App Structure
- apps/transcriber/apps/backend - NestJS API server (port 3006)
- apps/transcriber/apps/web - SvelteKit web application
- apps/transcriber/apps/landing - Astro marketing/content site
- apps/transcriber/apps/mobile - Expo React Native app
- apps/transcriber/packages/shared-types - Shared TypeScript types
## Backend Features
- YouTube video download via yt-dlp (child_process)
- Ultra-fast transcription via Groq Whisper API (~300x realtime)
- Fallback to local Whisper for offline use
- Job queue with background processing
- Real-time progress updates via WebSocket (Socket.io)
- Playlist management for batch processing
- Health check endpoints
## API Endpoints
- POST /transcription - Start transcription job
- GET /transcription - List all jobs
- GET /transcription/:id - Get job status
- DELETE /transcription/:id - Cancel job
- GET /transcription/stats - Statistics
- GET /whisper/models - Available models
- GET/POST/DELETE /playlist - Playlist management
- GET /health - Health checks
## Whisper Models
- Groq: whisper-large-v3-turbo (fast, $0.04/hr)
- Groq: whisper-large-v3 (accurate, $0.111/hr)
- Local: tiny, base, small, medium, large
## Monorepo Integration
- Added to pnpm workspace via apps/*/apps/* pattern
- Root scripts: transcriber:dev, dev:transcriber:*
- Package naming: @transcriber/{backend,web,landing,mobile}
- Turbo tasks: dev, build, lint, type-check
- CLAUDE.md documentation
## Technology Stack
- Backend: NestJS 10, TypeScript, Socket.io
- Web: SvelteKit 2, Svelte 5, Tailwind CSS
- Landing: Astro 4, Solid.js, Tailwind CSS
- Mobile: Expo 52, React Native, NativeWind, Zustand
- Transcription: Groq Whisper API (OpenAI-compatible)
## Migration from Python
- Original Python/FastAPI code preserved in legacy/
- Full rewrite to TypeScript/NestJS
- Same functionality with improved architecture
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
44 lines
No EOL
1.4 KiB
Bash
Executable file
44 lines
No EOL
1.4 KiB
Bash
Executable file
#!/bin/bash
|
|
# YouTube Transcriber - Schnellauswahl
|
|
|
|
source venv/bin/activate
|
|
|
|
echo "🎥 YouTube Transcriber - Modell-Auswahl"
|
|
echo "========================================"
|
|
echo ""
|
|
echo "1) 🚀 TINY - Schneller Test (39MB, ~10x Speed)"
|
|
echo "2) 🎯 LARGE - Beste Qualität (1.5GB, ~1x Speed)"
|
|
echo "3) 📋 SCAN - Alle Playlists scannen"
|
|
echo "4) ⚡ PARALLEL - Mehrere Videos parallel (3x Speed)"
|
|
echo ""
|
|
read -p "Wähle Modell (1-4): " choice
|
|
|
|
case $choice in
|
|
1)
|
|
echo "→ Nutze TINY Modell für schnellen Test"
|
|
read -p "YouTube URL: " url
|
|
python3 transcriber_v3.py process "$url" --model tiny
|
|
;;
|
|
2)
|
|
echo "→ Nutze LARGE Modell für beste Qualität"
|
|
read -p "YouTube URL: " url
|
|
python3 transcriber_v3.py process "$url" --model large
|
|
;;
|
|
3)
|
|
echo "→ Scanne alle Playlists mit LARGE Modell"
|
|
python3 transcriber_v3.py scan --model large
|
|
;;
|
|
4)
|
|
echo "→ Parallel-Verarbeitung (3x schneller!)"
|
|
echo "Gib URLs ein (mit Leerzeichen getrennt, oder Enter für Playlist):"
|
|
read -p "URLs: " urls
|
|
if [ -z "$urls" ]; then
|
|
python3 transcriber_v4_parallel.py process --playlist people/rory-sutherland --model large
|
|
else
|
|
python3 transcriber_v4_parallel.py process --urls $urls --model large
|
|
fi
|
|
;;
|
|
*)
|
|
echo "Ungültige Auswahl"
|
|
;;
|
|
esac |