ScribeCrate
Self-hosted speech-to-text API offering OpenAI-compatible transcription, translation, and speaker diarization.
Run faster-whisper locally with OpenAI API compatibility, speaker diarization, and offline support.
ScribeCrate is a self-hosted audio transcription server built on faster-whisper that mirrors OpenAI's Whisper API format. It allows hospitality operators to process recorded guest feedback, staff calls, or guided tour audio locally without routing audio to third-party cloud providers.
What it is
The server accepts audio files through REST endpoints to produce plain text, JSON, or SRT/VTT subtitle files. It supports foreign speech translation to English, speaker diarization via an optional sherpa-onnx extension, and progressive Server-Sent Events (SSE) streaming.
Who it's for
It suits hospitality software developers and IT engineers who need a local alternative to OpenAI's audio endpoints to retain data control and avoid per-minute API fees. It does not suit end users looking for a finished web dashboard or real-time microphone dictation out of the box.
Key features
OpenAI API compatibility
Provides standard endpoints for transcription and translation that match OpenAI Whisper API requests.
Speaker diarization
Identifies who spoke each audio segment using an optional local sherpa-onnx extension.
Server-Sent Events streaming
Streams decoded transcript segments progressively via SSE instead of waiting for full file processing.
Offline operation
Runs in isolated network environments using pre-cached Whisper models and local execution flags.
Up and running in 3 steps
Our honest verdict
What's great
- OpenAI-compatible REST API drops into existing software workflows with minimal code modifications.
- Supports local speaker diarization using sherpa-onnx to separate guest and staff dialogue.
- Provides SSE streaming so client applications display transcribed text progressively.
- Runs completely offline once speech models are downloaded and cached.
Where it falls short
- Requires container setup and server administration skills to deploy and maintain.
- Needs dedicated host memory and GPU hardware to run larger Whisper models efficiently.
- Lacks a user interface, requiring you to build or connect your own frontend.
- Requires configuring an external reverse proxy to secure API calls with HTTPS.
Pricing & license
ScribeCrate is free to use under the MIT licence. Your only costs are the underlying hardware host, GPU compute resources, and administration time required to run the server.
Open remote AI-training roles
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
Can ScribeCrate translate foreign guest audio into English?
Yes, it includes an OpenAI-compatible translation endpoint that converts foreign speech directly into English text output.
Does ScribeCrate include a web interface for hotel staff?
No, ScribeCrate operates strictly as a backend API service. You must connect it to an existing app or custom user interface.
What hardware do I need to host ScribeCrate?
You need a Linux server running Docker with at least 700 MB of free RAM for the base model, though an NVIDIA GPU is recommended for high-volume audio processing.
