Own the transcription pipeline
The product is designed to run speech recognition without depending on a third-party transcription API. That gives operators more control over data handling and deployment.
Independent product · Open source
Sonavra.
A self-hosted transcription workspace designed around privacy, resilient processing, and control over the infrastructure.
The challenge
Transcription should not require surrendering control of the workflow.
Uploading audio is the easy part. A dependable transcription product must handle long-running processing, failures, private storage, speaker-aware output, and the operational reality of running everything on your own infrastructure.
Engineering decisions
Designed for the difficult parts.
Treat processing as durable work
Uploads and transcription are separate operations. Persisted job states, retry behavior, and restart recovery are central to the architecture rather than an afterthought.
Keep files private
Audio belongs in private S3-compatible object storage. Presigned upload and controlled retrieval flows keep large files out of ordinary application request handling.
Make self-hosting practical
Local development and container packaging are part of the product. The goal is to make the full stack reproducible and straightforward to operate.
What this demonstrates
Product thinking, from interface to infrastructure.
Sonavra brings together user experience, API design, data modeling, asynchronous processing, object storage, local tooling, and deployment. It is an evolving engineering project, not a claim of production scale or commercial adoption.
Explore the source on GitHub ↗