Local-first meeting transcription with speaker matching and live captions.
Open-source, local-first meeting transcription and speaker diarization tool.
matrix-live-diarizer is a local-first application designed to convert meeting audio into structured, correctable transcripts without sending data to the cloud. It suits privacy-conscious professionals who need to manage meeting notes locally.
The tool processes audio through two main paths: post-meeting uploads with optional pyannote speaker separation, and real-time browser captions using Voice Activity Detection and ASR. It matches anonymous voices to registered speaker profiles, generates summaries using local Ollama or optional cloud LLMs, and exports to Markdown, SRT, VTT, or JSON.
It suits professionals who require strict audio data privacy and want to run transcription and diarization on their own hardware. It does not suit teams requiring public web deployment, multi-tenant setups, automated legal or medical identity authentication, or environments lacking technical capability.
Audio and transcripts remain on your machine with offline execution capability.
Matches anonymous speaker turns to registered individuals via file uploads or browser recordings.
Supports both post-meeting audio uploads and real-time browser captioning.
Allows switching between Qwen3-ASR, SenseVoice, and Paraformer runtimes.
Enables double-click transcript edits, bulk speaker reassignment, and summary generation.
The software is free under its MIT license. The only cost is the hardware resources, storage, and electricity required to run it yourself.
879 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
No. Audio processing and transcription happen entirely on your local machine by default.
An HF token is only needed if you want to use pyannote community-1 for multi-speaker diarization on uploaded files or utilize Qwen3-ForcedAligner for word-level timestamps.
Yes, once the initial models are downloaded during your first startup, the tool can run in an environment with no internet connection if LLM features are disabled.
See Zekai first in Google