How to Set Up Automatic Transcription on Your Own Server

Automatic transcription on your own server is less a feature you switch on than a small piece of infrastructure you plan: where recordings arrive, what the server does with each one, where the results go, and how long everything is kept. This guide walks through those decisions in the order you will meet them, so that installation becomes a matter of configuring choices you have already made.

It is written for whoever sets the pipeline up, usually IT together with the team that owns the meetings. If you are still deciding whether Scriber's automatic meeting transcription fits your organization at all, start with that page. This one assumes the answer is yes and deals with the setup.

Step 1: Decide where recordings will arrive

An automatic pipeline starts with a location the server watches. Scriber can pick up recordings from:

• A watched folder on the server. The simplest option when recordings can be copied straight onto the Scriber server, for example by an export job you already run.
• An S3-compatible bucket. A good fit if your organization already keeps files in S3-compatible object storage inside its own infrastructure.
• A mounted network share. The natural choice when recordings already end up on a shared drive, because nobody has to change where they save files.

The recordings themselves can come from your conferencing tool, phone system, or dictation device. What matters for the setup is not the source but the handover: find out where each of those currently writes its files, and pick the ingestion path closest to that place. If recordings already land on a network share, watching that share is usually less work than building a new copy step to the server.

Two practical checks belong here. First, the Scriber server has to be able to reach the location: the share mounted on the server, the bucket reachable from inside your network. Second, decide whether Scriber watches an existing location or a dedicated one. A dedicated folder or share makes it obvious which recordings are meant for transcription and keeps unrelated files out of the pipeline.

Step 2: Follow a recording from arrival to finished minutes

Once the location is watched, every new recording runs through the same sequence without anyone starting it:

1. Arrival. The file lands in the watched folder, bucket, or share, and Scriber picks it up.
2. Transcription. Our own transcription models produce a speaker-separated transcript on your server, so each statement is attributed to the person who made it.
3. Summary. Scriber writes an automatic summary of the discussion from that transcript.
4. Minutes. The finished minutes follow: title, date, attendees, and each agenda item with its discussion and decision, in the minutes template set up for your organization during installation.
5. Filing and delivery. The recording, transcript, summary, and minutes are filed in the searchable library and delivered to the destination you chose.

At no point in this sequence does a file leave the server or your storage. That is worth stating in your internal documentation, because it answers the first question most reviewers ask about an automated pipeline: where does the audio go while nobody is watching?

Step 3: Choose where the results are delivered

Results are delivered to the folder or network share you choose. The useful question is not which folder happens to be free, but where the people who need the minutes already work. If a department files its meeting documents on a particular share, deliver there, and the minutes appear next to the rest of that department's records.

Two things follow from that choice. Access rights come with the destination: whoever can read that folder can read the transcripts and minutes written to it, so pick a location whose permissions already match the confidentiality of the meetings. And everything delivered is also kept in the searchable library, so moving or renaming a delivered file later does not lose the record.

Then pick the formats. Scriber exports PDF, Word, plain text, Markdown, or JSON:

• PDF for minutes that are filed or circulated unchanged.
• Word for minutes that someone reviews and edits before they are approved.
• Plain text or Markdown for further processing, or for a wiki or documentation system.
• JSON for feeding transcripts and minutes into another internal system.

Step 4: Settle retention and storage before go-live

Because Scriber runs on your server and writes to your storage, retention, backup, and deletion are yours to set, and Determin never holds a copy. That is an advantage, but only if the rules exist before the first recording arrives. Settle these points in advance:

• How long recordings are kept. The original audio is often the most sensitive item. Decide whether it is kept as long as the minutes or removed sooner.
• How long transcripts and minutes are kept. Minutes may fall under a filing obligation; transcripts may not need to outlive the approved minutes.
• What goes into backups. Include the storage Scriber uses in the backup regime you already run, and make sure backup retention matches the retention you have agreed.
• Who may delete, and how. On a self-hosted installation, deletion means deleting from your own storage. Decide who handles deletion requests and how they are documented.

If recordings contain personal data, and meeting recordings nearly always do, agree these rules with your data protection officer. The article on the questions your DPO will ask covers retention, deletion, and the records of processing in more detail.

What to prepare before installation

Installation starts with an initial consultation. We then install and configure Scriber on your hardware, including the minutes template and the connection to the systems you use, and from then on operation runs entirely inside your network. The consultation goes faster if you bring the following:

• A Linux server with an Nvidia GPU, either in your network or rented from a GPU server provider.
• The ingestion path: the watched folder, bucket, or share Scriber should pick recordings up from, and which recordings belong there.
• The delivery destination and formats: where finished documents should appear and which formats their readers need.
• An example of your minutes: an existing set of minutes shows us what your template should look like.
• Retention, backup, and deletion rules, agreed between IT and your data protection officer.
• The network placement: Scriber works offline and on air-gapped networks, so decide whether it sits in your regular network or on an isolated segment. The page on on-premise transcription covers that infrastructure side in more depth.

If you want a first look before the consultation, a hosted demo runs at scriber.determin.de. It is a demo, not the product, so keep confidential recordings for your own installation. When your answers to the points above are ready, or when you want help working them out, get in touch.