ScribeCrate Live
Private, real-time speech-to-text
Open-source, self-hosted real-time transcription server.
Turn live microphone audio and audio streams into text on your own server. Powered by WhisperLive and faster-whisper, with WebSocket transcription and a separate OpenAI-compatible API for uploaded files.
Open source · MIT licensed
Try in Google Colab Stream sample audio in Google’s environment.
The Self-Hosted AI Builder's Guide Practical guidance for deployment, security, backups, and everyday operation.
Live audio → progressive text
A microphone or audio stream, sent by your client
Let’s review the deployment plan.
A focused service for your applications
Build live captions, microphone transcription, and applications that consume progressive text. You operate the server and supply the client that captures or streams audio.
Connect a compatible client to port 9090 for progressive transcription of microphone audio or audio streams. Follow the documented handshake and audio format.
Use POST /v1/audio/transcriptions on the REST service at port 8000 for compatible file-transcription clients. WebSocket and REST serve different input flows.
Transcription runs on your server. Fresh persistent installations enable API-key authentication for transcription, with a generated key stored across restarts.
Choose a supported Whisper model and use voice activity detection (VAD) to skip silence. Larger models can require substantially more memory and processing time.
CPU images support amd64 and arm64. Use the separate :cuda image for NVIDIA acceleration on amd64 with a compatible driver and NVIDIA Container Toolkit.
Keep downloaded models across restarts. For offline operation, cache the required models first and enable WHISPERLIVE_LOCAL_ONLY.
Images are automatically built and published through public GitHub Actions workflows. Powered by WhisperLive, faster-whisper.
Start with a local server
Prefer to try it first? Open the Google Colab demo. Audio submitted there is processed in Google’s environment.
This CPU example binds both services to loopback. The REST request gives you an authenticated first test. The Colab demo streams recorded audio; use the documented WebSocket client for microphone input.
For access from other computers, follow the maintained HTTPS, authentication, and network-exposure instructions.
docker run \
--name whisper-live \
--restart=always \
-v whisper-live-data:/var/lib/whisper-live \
-p 127.0.0.1:9090:9090 \
-p 127.0.0.1:8000:8000 \
-d hwdsl2/whisper-live-serverCheck docker logs whisper-live for readiness. The default base model downloads on first client connection.
scribe_live_api_key="$(docker exec whisper-live whisper_live_manage --getkey)"Fresh persistent installations generate and save a key. REST transcription uses a Bearer header; WebSocket clients use the documented token connection flow.
curl http://127.0.0.1:8000/v1/audio/transcriptions \
-H "Authorization: Bearer $scribe_live_api_key" \
-F file=@audio.mp3 \
-F model=whisper-1Replace audio.mp3 with your file and run in the same shell. Live microphone capture requires a compatible WebSocket client; the server does not capture your microphone by itself.
Part of a larger toolkit
ScribeCrate Live is also available as part of Self-Hosted AI Stack. Deploy local models, chat, document processing, voice, and AI tools together, or choose a smaller configuration for your needs.
Explore Self-Hosted AI StackConnect the services you need
ScribeCrate provides an API for uploaded recordings, subtitles, English translation, and optional local speaker diarization. Choose the service that fits your input flow.
Explore ScribeCrateThe ScribeCrate Live project is open source under the MIT License. You provide hardware or cloud hosting and pay associated costs. Models, dependencies, and external providers retain their own licenses and terms.
No. This is an informational website. A client you run captures audio and sends it to your deployment. The optional Google Colab demo runs in Google’s environment and follows its data practices.
ScribeCrate focuses on uploaded-file transcription, subtitles, English translation, and optional speaker diarization. ScribeCrate Live focuses on live WebSocket transcription and also provides a REST endpoint for uploaded-file transcription. They are separately deployed projects.
The REST service provides an OpenAI-compatible transcription endpoint for supported file clients. Live audio uses the WhisperLive WebSocket protocol. Do not assume support for every OpenAI audio endpoint, parameter, or realtime protocol.
Cache the required Whisper models before disconnecting and set WHISPERLIVE_LOCAL_ONLY=true. Disable optional aggregate usage counts separately with WHISPERLIVE_DISABLE_USAGE_COUNTS=1.
Use the maintained WebSocket client instructions for microphone capture. The Google Colab demo demonstrates streaming a prerecorded file. Client authentication, handshake fields, and audio encoding must match the server. Use HTTPS/WSS and authentication for remote connections.
The image includes optional aggregate usage counts.
See the usage-count documentation;
set WHISPERLIVE_DISABLE_USAGE_COUNTS=1 to disable them.
Bring your live audio. Keep control of the transcription server.
View on GitHub