Whisper Speech-to-Text on Docker
A Docker image that runs a self-hosted Whisper speech-to-text server with an OpenAI-compatible transcription and translation API, optional speaker diarization, and GPU acceleration.
This image runs faster-whisper behind endpoints that match OpenAI's own audio transcription and translation APIs, so an application built against OpenAI's Whisper API can switch to it with a one-line base-URL change. It supports every published Whisper model size, JSON, plain-text, SRT and WebVTT output formats, server-sent-events streaming so a client receives segments as they are decoded rather than waiting for the whole file, and an offline mode that runs from pre-cached models without internet access.
Optional local speaker diarization identifies who is speaking in each segment, using a separate on-device extension. A companion image instead targets near-real-time transcription from a live microphone or an RTSP stream over WebSocket, while this image is built for complete audio files over HTTP.
It supports NVIDIA CUDA acceleration and multiple CPU architectures, and is built on Debian.