WhisperLiveKit

Why we included this project

WhisperLiveKit covers the case where batch transcription won't do: when you need text as people are speaking, not after the recording ends. Standard Whisper expects a complete utterance, so feeding it live audio chunks comes back as cut-off words and missing context; this project applies simultaneous-speech research to buffer incoming audio and stream the transcript as it goes. That makes it useful for live captioning, meeting transcripts, and other realtime workloads where the text has to arrive with the speech. It also bundles real-time speaker diarization and translation into about two hundred languages, and it exposes OpenAI- and Deepgram-compatible endpoints, so existing clients can point at a local server without rewriting. Since the whole pipeline runs on your own hardware, you keep the audio in-house and avoid per-minute API fees.