Can I use it as a drop-in for the OpenAI Whisper API?

Yes. The endpoint is OpenAI-compatible: any library or integration that targets the OpenAI transcription API can be redirected to your private instance simply by changing the base URL. No API key is required on the server side — access is managed by the reverse proxy.

Do my audio files leave the VPS?

No. The audio is processed entirely locally on your VPS. No data is sent to OpenAI or any third party. Files are kept in memory during processing and are not stored on disk after the transcription is returned.

Build, host and run AI solutions.

Whisper

Private speech-to-text API on your VPS — multilingual, fast, OpenAI-compatible.

1 GB minimum (base model), 2 GB recommended (small model) RAM 1 vCPU In validation

In validation

Automatic installation for this solution is ready and currently going through our tests on a real server. Ordering will open as soon as validation is complete.

Whisper is OpenAI's open-source speech recognition model, here deployed as a self-hosted REST API via the faster-whisper CTranslate2 backend. Running on your VPS, it transcribes audio files in 99 languages, translates to English, and returns word-level timestamps — all without sending a single byte to a third-party service. The `base` model runs comfortably on a 1-vCPU VPS with 1 GB of RAM; the `small` model lifts accuracy further with 2 GB. Use it to power captioning, search indexing, meeting notes, podcast processing or any workflow that converts voice to text.

Key features

OpenAI-compatible REST API: POST /asr, output as txt, json, srt, vtt or tsv.

faster-whisper CTranslate2 backend: up to 4× faster than original Whisper on CPU.

99 languages supported — Arabic, French, English, Darija and more — in a single model.

Word-level and segment-level timestamps, ready for subtitle generation.

No per-minute billing and no data leaving your VPS.

Model choice via ASR_MODEL env var: tiny, base, small, medium, large-v3.

When to use this solution?

1

Meeting and interview transcription

2

Podcast and media subtitling

3

Voice-to-text for internal tools

Deploy Whisper on your VPS

Guide optimized for ServOrbit Cloud VPS.

01

Deploy from the Marketplace

Open the ServOrbit dashboard, go to Marketplace → Artificial Intelligence → Whisper, and click Deploy. Docker pulls the image and starts the container in under two minutes. The first start also downloads the ASR model weights (~270 MB for base); the download is cached so subsequent restarts are instant.

02

Access the API

After installation, use the address shown in your client area. The Swagger UI is available at /docs for interactive testing. The main endpoint is POST /asr: attach your audio file and set the output parameter to txt, json, srt, vtt or tsv according to the format you need.

03

Transcribe your first file

From a terminal or your application, send a request: ` curl -F "[email protected]" "https://your-address/asr?output=txt"You will receive the full transcription as plain text. For timestamped segments addoutput=json`.

04

Change the model if needed

The default model is base (fast, ~140 MB). To improve accuracy, switch to small (better for noisy audio, ~500 MB) by updating the ASR_MODEL environment variable. A VPS with 2 GB of RAM handles small comfortably. Contact support if you want to resize your VPS.

05

Integrate into your pipeline

The endpoint is OpenAI Whisper-compatible: any library or tool that can call the OpenAI transcription API can be pointed at your private endpoint instead. Update the base URL and remove any API key — authentication is handled at the reverse proxy level by ServOrbit.

Frequently asked questions

The model recognises 99 languages including Arabic, French, English, Spanish, Darija and many more. Transcription and translation to English are handled by the same model, selected via the task parameter (transcribe or translate).

The base model (default) runs comfortably on 1 GB of RAM and 1 vCPU. It delivers solid accuracy for clear speech. For noisy recordings or lower-resource languages, small is a better fit and works well on a 2 GB VPS. The large-v3 model requires 8 GB of RAM and is recommended for GPU VPS plans.

Whisper is deployed as a REST API with a Swagger UI at /docs for interactive testing. It does not include a standalone file upload web app. You can call the API from your own interface, from tools such as n8n or Make, or directly from the command line with curl.

Complementary solutions

Artificial Intelligence

Open WebUI

Web interface to interact with your local or remote LLMs. Your data stays on your infrastructure — no third party involved.