WhisperLive

Use it when

• You need live partial transcripts from a microphone, RTSP camera, or HLS broadcast stream instead of waiting for a whole file to finish processing
• You must keep audio on your own hardware and want a self-hosted Whisper server with word timestamps, hotword boosting, and optional speaker diarization

What it solves

Not the fit when

• Managed cloud-scale transcription; you operate the server and compute
• Exposing the WebSocket or REST service to untrusted networks without a separate authentication, TLS, and access-control boundary
• Assuming speaker labels without installing and enabling the optional `pyannote.audio` diarization dependency
• TensorRT deployment without first building the required engines; Docker is the documented recommended path
• text to speech synthesis
• llm chat inference
• document ocr
• video frame analysis

Install

pip install whisper-live (after bash scripts/setup.sh for PortAudio), or Docker: docker run -it -p 9090:9090 ghcr.io/collabora/whisperlive-cpu:latest (GPU: whisperlive-gpu)

Invoke

Server: python3 run_server.py --port 9090 --backend faster_whisper; client: from whisper_live.client import TranscriptionClient; client = TranscriptionClient("localhost", 9090, lang="en", model="small"); client() for microphone or client("audio.wav") for files

Alternatives

No reviewed alternatives recorded yet.