Skip to content

Understanding transcription

What Is Self-Hosted Transcription?

Self-hosted transcription runs speech-to-text on hardware you control, so recordings stay with you. What it is, what it costs, and when it is worth it.

The Scriptorium team ·

Self-hosted transcription is speech-to-text software that you run on your own computer or server, instead of uploading recordings to a company's cloud. The audio, the transcript and the search index all stay on machines you administer.

How it differs from a cloud service

With a cloud transcription service the recording travels to the vendor, is processed on their infrastructure and the result is stored in their account system. That is convenient, and for most recordings it is perfectly reasonable. With self-hosting, the same work happens beside the files. Nothing leaves unless you send it.

What actually runs on your machine

Modern open speech models, the Whisper family in particular, are small enough to run on ordinary hardware. The application loads a model from disk, decodes the recording and produces timestamped text. The one thing that needs a network is the first download of the model itself; after that, transcription needs none.

Who it is for

  • Lawyers, clinicians and researchers whose recordings carry a confidentiality duty.
  • Organisations whose contracts forbid passing recordings to third-party processors.
  • Teams that want control over retention, access and sign-in.
  • Anyone who simply prefers their conversations stay on their own disk.

What it costs you

Self-hosting trades convenience for control, and the cost is real. You are the administrator: you install and update it, back up the data, put it behind HTTPS if other people will use it and decide who has access. Transcription also uses your own processor, so it is slower on modest hardware than on a large cloud machine, and bigger models need more memory.

Self-hosting is a deployment choice, not a guarantee. A badly configured server exposed to the internet is less private than a reputable cloud service. The privacy benefit comes from running it carefully.

Hosted and self-hosted are different things

The public site at tryscriptorium.com is a hosted service: a file you upload is processed on its server. That is fine for a non-sensitive recording. For confidential material, self-hosted transcription on your own hardware is the option that keeps the recording under your control.

Questions

Is self-hosted transcription more accurate than cloud transcription?
Not inherently. Accuracy depends on the speech model and the quality of the recording. The same open models can run either way; self-hosting changes where the work happens, not how well the model hears.
Do I need a GPU?
No. Transcription runs on an ordinary processor. It is slower than on a GPU, and larger models need more memory, so pick a model size that suits your hardware.
Does it need an internet connection?
Only to download the speech model the first time. After that, transcription runs from the model stored on disk.
Is it free?
That depends on the software. Running it yourself removes per-minute fees, but you provide the hardware and the administration.

Transcribe a file

Free for a first file, with no account.