whisper.cpp Speech to Text Server

This is a repackaged open source software product wherein additional charges apply for cloudimg support services.

## What It Does

whisper.cpp is a high-performance C/C++ port of OpenAI's Whisper automatic speech recognition model, running entirely on the CPU with no GPU, no Python runtime, and no external service dependencies. This image delivers it fully built and configured with a speech-recognition model already installed and an HTTP API and browser interface operational within minutes of launch. Transcribe audio without compiling anything or hunting for model files.

## Why This Image Instead of a DIY Install

• Versus a manual EC2 setup: Skip cloning and compiling whisper.cpp, choosing and verifying a model file, writing a systemd unit, and standing up an authenticating reverse proxy. This image handles all of that at launch.
• Versus exposing the bare server: The upstream server has no authentication and a remote model-swap primitive; this image binds it to loopback, fronts it with nginx basic auth on port 80, and blocks the model-swap endpoint outright.
• Versus other community images: A hardened first-boot process generates a unique password per instance, the MIT licence position is documented with a written provenance record, and 24/7 cloudimg support is included.

## Security - Hardened From First Boot

• Secure first boot: A one-shot service generates a fresh password per instance and writes it to a root-only file before the endpoint is reachable. No shared or default credentials ship in the image.
• Network isolation: The inference engine binds to loopback only. The single public surface is nginx on port 80 (basic auth) and SSH on port 22. Deploy in a private VPC subnet and use security groups to restrict access.
• Encryption at rest: Enable AWS EBS encryption when launching the instance, or use encrypted snapshots.
• Blocked model-swap endpoint: The upstream endpoint that can load an arbitrary server-side file as a model is blocked at the reverse proxy layer, eliminating a documented attack surface.

## Use Cases

• Media production and podcasting: Batch-transcribe episodes and generate SRT/VTT subtitle files for publishing workflows, keeping all audio data on your own infrastructure.
• Contact center and compliance teams: Process recorded calls on a private, self-hosted API with no data sent to a third-party service - ideal for organizations with data residency requirements.
• Developer pipelines: Feed transcription output (plain text, JSON, SRT, or VTT) into downstream NLP, search indexing, or compliance archival systems via the HTTP API.

## How It Works

Transcribe by posting an audio file to the /inference endpoint or by uploading it in the browser interface. Many audio and video formats are accepted and converted automatically. Results are available as plain text, JSON, SRT, or VTT subtitles.

The preinstalled model is a format conversion of OpenAI's original Whisper base English checkpoint. Both the model weights and whisper.cpp itself are distributed under the MIT License, and copies of both licences together with a written provenance record ship in the image, so the licensing position is documented rather than assumed.

## Getting Started

1. Launch the image and open port 80 (and optionally port 22) in your security group.

2. SSH into the instance and retrieve the generated credentials from /root/whisper-cpp-server-credentials.txt.

3. Browse to the instance public IP to reach the browser UI, or POST audio files to /inference using the credentials.

For production architecture planning or instance sizing recommendations for your workload, contact cloudimg support at any time.

## 24/7 cloudimg Support

Technical support by email and live chat covers deployment, instance sizing for CPU inference, credential management, model provenance questions, and API integration. Critical issues receive a one-hour average response time.

Whisper is a trademark of its respective owner. All product and company names are trademarks or registered trademarks of their respective holders. Use of them does not imply any affiliation with or endorsement by them.