WhisperLiveKit is an open-source toolkit for real-time, fully local speech-to-text. Built on OpenAI’s Whisper models, it transcribes live audio streams with ultra-low latency, labels who is speaking, and can translate speech on the fly — all on your own hardware, with nothing sent to the cloud. It has earned about 11,200 stars on GitHub and was actively developed through 2026.
The kit wraps Whisper and faster-whisper in a streaming pipeline with voice activity detection, buffered incremental decoding and speaker diarization powered by Diart. Audio arrives over WebSocket or from files and URLs, and partial transcripts stream back word by word. For integration, it exposes OpenAI- and Deepgram-compatible API endpoints, so existing transcription clients can point at your self-hosted server unchanged. A FastAPI backend, multi-user handling and GPU acceleration round it out, making it practical for live captions, meeting transcription and voice interfaces that must stay private and on-premises. Recent releases have focused on multilingual transcription, word-level timestamps and restored punctuation, making the output usable for subtitling as well as live captioning. The server can transcribe several simultaneous streams on a single GPU, and a browser-based demo interface is included for quick tests. Everything installs as a single Python package, with configuration for model size, language, buffering and diarization so you can trade latency against accuracy for your hardware.
WhisperLiveKit is free and open source under the Apache 2.0 license. Your only cost is the hardware you run it on.
WhisperLiveKit is for developers and teams who need live transcription — captions, meetings, voicebots, accessibility tools — without sending sensitive audio to a third-party cloud API.