Transcription Stream is a turnkey, self-hosted transcription and speaker diarization service that works completely offline, with 946 stars on GitHub.
Transcription Stream packages speech-to-text into a service you can run on your own infrastructure. It combines Whisper-based transcription with speaker diarization, so you not only get the words but also who said them. Everything works offline: audio never has to leave your network, which makes it suitable for sensitive recordings such as interviews, legal material or internal meetings. A web interface lets you upload files, listen back, review transcripts and download results, while SSH drop zones offer a drag-and-drop path for automated workflows. Once transcribed, built-in LLM summarization through Ollama and Mistral models turns long recordings into readable summaries, and Meilisearch provides fast full-text search across your archive. Because the whole stack is self-contained and container-friendly, a single modest server can cover a small team’s transcription needs, and finished transcripts stay searchable in the archive long after the original audio was processed.
Transcription Stream’s Community Edition is free and open source under the GPL-3.0 license. You self-host it on your own hardware and pay nothing per minute — a sharp contrast with cloud transcription services that charge for every hour of audio.
Transcription Stream is for teams and individuals who transcribe regularly but cannot or will not send audio to the cloud — journalists, researchers, podcasters, legal and healthcare-adjacent workflows, and privacy-conscious organisations. If you have a machine to run it on, it replaces a recurring transcription bill with a one-time setup.