ttlequals0/minuspod
Removes ads from podcasts using Whisper + AI. Serves clean RSS feeds for any podcast app.
Image
Machine learning & AI
1
100K+
ttlequals0/minuspod repository overview
MinusPod
Removes ads from podcasts using Whisper transcription and AI detection. Serves modified RSS feeds that work with any podcast app.
How It Works
- Transcription -- Whisper converts audio to text with timestamps (local GPU or remote API)
- Ad Detection -- 3-stage pipeline: audio fingerprinting, learned text patterns, then LLM analysis (Claude, Ollama, or any OpenAI-compatible endpoint)
- Audio Processing -- FFmpeg removes detected ads with short audio markers
- Serving -- Flask serves modified RSS feeds and processed audio
Processing is on-demand. First play takes a few minutes; subsequent plays are instant (cached).
Features
- Web UI for feed management, ad review/correction, pattern management, and settings
- Ad editor with boundary adjustment, confidence scores, and detection stage badges
- Pattern learning system (podcast, network, and global scope) -- improves over time
- Verification pass re-transcribes processed audio to catch missed ads
- Audio analysis (volume anomalies, transition detection, DAI boundary enforcement)
- Webhooks for notifications (Pushover, ntfy, or any HTTP endpoint)
- OPML import/export, full-text search, processing history with stats
- REST API with 71 documented endpoints (OpenAPI spec at
/docs)
Quick Start
cat > .env << EOF
ANTHROPIC_API_KEY=your-key-here
BASE_URL=http://localhost:8000
EOF
mkdir -p data
docker-compose up -d
Web UI at http://localhost:8000/ui/
Requirements
- NVIDIA GPU with Docker support (for local Whisper), or use a remote Whisper API
- Anthropic API key, Ollama, or any OpenAI-compatible LLM endpoint
LLM Providers
| Provider | Config |
|---|---|
| Anthropic (default) | ANTHROPIC_API_KEY=your-key |
| Ollama (local, free) | LLM_PROVIDER=ollama, OPENAI_BASE_URL=http://host.docker.internal:11434/v1, OPENAI_MODEL=qwen3:14b |
| OpenAI-compatible | LLM_PROVIDER=openai-compatible, OPENAI_BASE_URL=http://your-endpoint/v1 |
Whisper Backends
| Backend | Config |
|---|---|
| Local GPU (default) | WHISPER_MODEL=small, WHISPER_DEVICE=cuda |
| Remote API | WHISPER_BACKEND=openai-api, WHISPER_API_BASE_URL=http://your-server:8765/v1 |
| Groq | WHISPER_BACKEND=openai-api, WHISPER_API_BASE_URL=https://api.groq.com/openai/v1 |
Volumes
| Path | Description |
|---|---|
/app/data | Database, cached RSS, and processed audio |
/app/assets | (Optional) Custom replace.mp3 for ad break marker |
Environment Variables
| Variable | Default | Description |
|---|---|---|
ANTHROPIC_API_KEY | -- | Claude API key |
LLM_PROVIDER | anthropic | anthropic, openai-compatible, or ollama |
OPENAI_BASE_URL | -- | Base URL for OpenAI-compatible/Ollama |
OPENAI_API_KEY | not-needed | API key for OpenAI-compatible endpoint |
OPENAI_MODEL | -- | Model name for non-Anthropic providers |
BASE_URL | http://localhost:8000 | Public URL for feed links |
WHISPER_MODEL | small | Whisper model size |
WHISPER_DEVICE | cuda | cuda or cpu |
WHISPER_BACKEND | local | local or openai-api |
WHISPER_API_BASE_URL | -- | Remote Whisper API URL |
TUNNEL_TOKEN | -- | Cloudflare tunnel token |
Links
Disclaimer
This tool is for personal use only. Only use it with podcasts you have permission to modify or where such modification is permitted under applicable laws.
Tag summary
Content type
Image
Digest
sha256:246004689…
Size
4.4 GB
Last updated
3 days ago
docker pull ttlequals0/minuspod