learnedmachine

As is mentioned above, you do not need to use cloud services; you can host your own. Faster whisper or WhisperX based server for trasncription and any openAI compatible API server (ollama, vllm, sglang, there are so many more) for LLM inference. Googling and reading the docs will help.

Speakr - Self-Hosted AI Transcription Platform Docker Hub: https://hub.docker.com/r/learnedmachine/speakr GitHub: https://github.com/murtaza-nasir/speakr Documentation: https://murtaza-nasir.github.io/speakr Overview Speakr transforms your audio recordings into organized, searchable, and intelligent notes. Built for privacy-conscious individuals and groups, it runs entirely on your own infrastructure. Key Features • 🎙️ AI-Powered Transcription - High-accuracy transcription with speaker identification • 🗣️ Voice Profiles - Automatic speaker recognition using voice embeddings • 💬 Interactive Chat - Ask questions about your recordings • 🔍 Semantic Search - Search across all recordings using natural language (Inquire Mode) • 🤝 Collaboration - Share recordings with users and groups • 🏷️ Smart Tagging - Custom AI prompts and ASR settings per tag • 📤 Auto-Export - Export to Obsidian, Logseq, and other note-taking apps • 🌍 Internationalization - Full support for EN, ES, FR, DE, ZH • 🎨 Themes - Light/dark modes with customizable color schemes Installation on Unraid Prerequisites You’ll need API keys for: • Text Generation: OpenRouter (recommended) or OpenAI • Transcription: OpenAI Whisper API OR you can host your own services for ASR and LLM inferencing. Get your API keys: • OpenRouter: https://openrouter.ai/keys (supports many models including GPT-4o-mini) • OpenAI: https://platform.openai.com/api-keys Quick Start 1. Install from Community Applications - search for “Speakr” 2. Configure required fields: – Text Model API Key - Your OpenRouter or OpenAI key – Transcription API Key - Your OpenAI Whisper key – Admin Password - Choose a secure password 3. Click Apply 4. Access at http://YOUR-UNRAID-IP:8899 5. Log in with username admin and your chosen password Default Configuration The template uses sensible defaults: • Port: 8899 • Storage: /mnt/user/appdata/speakr/ • Text Model: OpenRouter with GPT-4o-mini (cheap and fast) • Registration: Disabled (only admin can create accounts) Advanced Features Speaker Diarization (Optional) For advanced speaker identification with voice profiles, you can run a separate ASR service: • WhisperX ASR Service (recommended): https://github.com/murtaza-nasir/whisperx-asr-service • Basic ASR Service: https://github.com/ahmetoner/whisper-asr-webservice Then update the environment variables to use your ASR endpoint instead of OpenAI. These services will need a CUDA capable GPU. Auto-Processing Enable the “Auto-Process” feature to automatically transcribe files dropped into a watch directory: 1. Set ENABLE_AUTO_PROCESSING=true 2. Drop audio files into /mnt/user/appdata/speakr/auto-process/ 3. Files are automatically processed and moved to your upload folder Auto-Export to Obsidian/Logseq Set ENABLE_AUTO_EXPORT=true and point the exports volume to your note-taking app’s import folder. Common Use Cases • Meeting notes - Record and transcribe team meetings with speaker identification • Interviews - Transcribe research interviews with automatic export • Lectures - Convert lecture recordings to searchable notes • Podcasts - Generate transcripts and show notes • Legal/Medical - Securely transcribe consultations (all data stays on your server) Documentation Complete documentation available at: https://murtaza-nasir.github.io/speakr Support • Issues/Bugs: https://github.com/murtaza-nasir/speakr/issues • Questions/Discussion: https://github.com/murtaza-nasir/speakr/discussions • Unraid Support: This thread Screenshots Screenshots: https://murtaza-nasir.github.io/speakr/screenshots License Dual-licensed under AGPLv3 (open source) and commercial license. ________________________________________ Post questions and issues in this thread. I’ll do my best to help!