Digital sovereignty in ASR: where the data actually lives
Where do GoodTape, Speechmatics, HappyScribe, AmberScript and Scriptix actually process your audio? And which of them let you run the engine inside your own walls?
For a growing list of buyers, residency and deployment are no longer 'nice to have' clauses negotiated at the end of a contract. They are gating criteria. Procurement teams are reading the fine print, asking where keys are held, who has operator access, and whether the engine can run with no outbound network at all. We mapped where the main ASR vendors actually run, and which of them offer on premise or air gapped options.
Where each vendor processes audio
| Vendor | Primary processing region | On premise | Air gapped |
|---|---|---|---|
| Scriptix | EU (Netherlands) by default | Yes, licence key | Yes, on request |
| GoodTape | EU (Denmark) | No | No |
| Speechmatics | UK / multi region | Yes (container) | Partial, vendor managed |
| HappyScribe | EU and US (engine dependent) | No | No |
| AmberScript | EU (Netherlands) | No (SaaS) | No |
Most of the well known vendors keep European processing, which is enough for many GDPR conversations. But residency is only part of the picture. If a vendor still controls the keys, the operating system and the network egress, the audio is in the EU but it is not in your hands.
Three levels of sovereignty
It helps to be explicit about what 'sovereign' actually means, because vendors use the word very loosely. There are three useful levels:
- Residency. The data is processed and stored in a specific region. Most EU SaaS vendors offer this.
- On premise. The engine runs inside the customer's infrastructure. The vendor cannot read the audio, but typically retains operational responsibilities (updates, telemetry, licence checks).
- Air gapped. The engine runs with no outbound network. No callbacks, no telemetry, no licence phone home. Updates arrive through controlled artefacts.
On premise is the meaningful test
An on premise deployment moves the engine inside the customer's own infrastructure: their datacentre, their Kubernetes cluster, their virtual private cloud. It removes the vendor from the data path. For ministries, courts and intelligence services, that is the line between 'GDPR compliant' and 'sovereign'.
Of the vendors above, only Scriptix and Speechmatics support a real on premise install, and both rely on licence keys to activate the deployment. With Scriptix, a partner requests a key for ASR only or the full platform, we grant it from our cloud backoffice, and the customer deploys in their own tenant. HappyScribe, GoodTape and AmberScript are SaaS only.
Air gapped is the sovereign test
An air gapped deployment goes one step further: no outbound network access, no callbacks, no telemetry. Scriptix supports this on request. It is not the default install path because the contract, setup and update process all look different, so it always starts with a conversation.
Speechmatics offers an on premise container, but the operational model still leans on vendor managed updates and observability. That is not a flaw, it is a design choice. It just means the air gapped tier is harder to reach without bespoke engineering on the customer side.
Why this matters now
The Data Governance Act, NIS2, DORA in financial services, and procurement guidance from EU institutions are all pushing in the same direction: critical workloads should run on infrastructure the buyer controls. ASR has historically been a SaaS first market because the models are too large and the compute is too expensive to ship to every customer. That is changing, and the vendors that designed for sovereignty from day one have a structural advantage as the regulatory floor rises.
If sovereignty matters to your organisation, the question to ask any ASR vendor during evaluation is simple. Can you run with no outbound network, and can you prove it? The answer separates marketing language from product reality very quickly.
About the author
Frans Olsthoorn founded Scriptix in 2010 and has spent more than fifteen years shipping speech recognition into European broadcasters, courts and government bodies. He writes about ASR, accessibility regulation and the realities of running AI workloads inside customer infrastructure.
Connect on LinkedIn