WhisperLive Real-Time Speech-to-Text on Docker
A Docker image for a self-hosted, real-time speech-to-text server based on faster-whisper, offering WebSocket streaming transcription and an OpenAI-compatible REST API, with optional GPU acceleration.
docker-whisper-live packages a self-hosted WhisperLive server, powered by faster-whisper, as a Docker image built on a Debian-based Python 3.12 slim base. It provides WebSocket streaming for transcribing live microphone audio or audio streams in near real time, and an OpenAI-compatible REST endpoint for transcribing complete files, so an application built against the OpenAI Whisper API can switch to it with a one-line change.
It supports all standard Whisper model sizes, voice activity detection to skip silence for faster and cleaner transcription, a helper script for managing installed models, NVIDIA GPU acceleration through a separate CUDA image tag, and an offline mode that runs entirely from pre-cached models without internet access. Audio stays on the server rather than being sent to a third party, images are built and published automatically through GitHub Actions, and both amd64 and arm64 architectures are supported.
It is part of a broader self-hosted AI stack the same maintainer publishes alongside related images for Whisper, embeddings and other services.