Make your recordings searchable.
To you. To no one else.
Switzerland is becoming a surveillance jurisdiction. Austria is the alternative. The Swiss VÜPF revision pending in 2026 would mandate provider-side decryption, ID checks at 5,000 users, and 6-month log retention — stricter than the EU or the U.S. Proton, the Swiss privacy flagship, is publicly relocating infrastructure out of Switzerland. SoTranscribe is built directly in Austria: open-source pipeline, Anexia Klagenfurt hosting, no US providers in the chain.
Open source under the EvoBioSys License. No credit card. No tracking.
Why this exists
Your audio is yours. The leverage shouldn’t be theirs.
When you upload audio to a cloud transcription service, you’re trusting a long chain of companies with what was said in your meeting, your therapy session, your interview, or your private voice note. The audio gets copied, stored, processed, and sometimes used to train someone else’s product.
Most of the popular tools are American companies. A US court can request your data even when it’s “stored in Europe.” There is a friendlier way to do this — and we’re building it.
Jurisdiction matters
Why Austria.
A year ago we would have said “Austria or Switzerland, you pick.” In 2026 the answer is simpler. Austria is in materially better jurisdictional shape than Switzerland — and even Switzerland’s flagship privacy company is moving infrastructure out.
Austria — protections in place
EU member · GDPR · court-order regime
- § 134 / 135a StPO requires an individual court order before any provider can be compelled to release content data
- Verfassungsgerichtshof, 2014 struck down general data-retention obligations as unconstitutional — no bulk IP-log mandate
- EU and GDPR-aligned: a single, well-tested legal regime, not a moving target
- Hosting on Anexia (Klagenfurt & Wien) — Austrian-owned, named facility
Switzerland — what’s under threat
Pending VÜPF revision · Federal Council ordinance
- Pending VÜPF revision (Article 50a) would mandate provider-side decryption — a structural backdoor
- Data-retention obligation kicks in at just 5,000 users (6 months of IP logs)
- Government-ID identification mandate for covered Swiss services
- Being implemented via Federal Council ordinance — bypasses parliamentary debate
- Proton, the Swiss privacy flagship, is publicly relocating infrastructure out of Switzerland into the EU
Proton ist dabei, die Schweiz zu verlassen. Wir bauen direkt in Österreich.
We’re not puritans. You don’t have to self-host. You don’t have to give up the convenience. You just stop bleeding meeting audio into someone else’s ad system — and you don’t bet on a jurisdiction that’s mid-transition.
What we promise
Three things, in plain language.
Your audio stays where you put it.
Run it on your own computer (free) and nothing leaves the room. Use our hosted tier and audio sits on Austrian servers we own, never shared with anyone.
No US company touches your data.
We don’t use Stripe, Google Analytics, or American clouds. Mollie (NL) for payments, Bunny Fonts (SI), Plausible (EU) if we ever add analytics. So a US court can’t reach in.
You can leave any time.
Your transcripts are yours, in standard formats. Cancel and you take everything with you. The source code is open, so the door always opens outward.
How it works
Three steps. One result.
-
A
Drop in your recordings.
Voice memos, meetings, podcast episodes, wearable captures — whatever you have. Or set a folder to auto-import as new ones land.
-
B
We transcribe and tell speakers apart.
Each conversation becomes a clean text document with timestamps and who-said-what. Whisper Large v3 + pyannote diarization. No human listens. No third-party service involved.
-
C
Search like you search email.
Press ⌘F — or open the app — and find every moment where you said the word you’re looking for. Across years of conversations.
What people use it for
Some of the things it’s good at.
Pricing
Pick what feels right.
Four ways to use SoTranscribe, plus a custom enterprise option. Self-host stays free, forever. Hosted tiers are still in shaping — final pricing will be set transparently via Open Collective Europe before Beta launches.
Self-host
Free
Run it on your own computer. Source open under the EvoBioSys License. A laptop with 8 GB Apple Silicon or an RTX-3060-class GPU is enough.
Get the code →Most frictionless
Hosted (Light / Plus)
€12–€25/mo
We host it for you on Austrian servers (Anexia, Klagenfurt & Wien). 100% renewable energy. Or pay €8/mo with the Patient Green tier and queue your transcripts for sun + wind peaks.
Join the waitlist →For organisations
Enterprise
Custom
Larger volume, custom integration, customer support, signed Auftragsverarbeitungsvertrag. Hosted in Austria, GDPR-native, no US providers in the chain.
Talk to us →Patient Green · €8/month
For users who don’t need instant turnaround. Queue your transcripts and they’ll process when our wind + solar peaks. Could be 2–48 hours. 30%+ off the standard rate. The most ecologically honest option.
How we compare
Apples-to-apples: cost per minute, jurisdiction, openness, energy.
Subscription pricing varies wildly by quota, and "$17/mo" without included minutes hides what you actually pay. Below: cost normalised to per minute of transcribed audio, plus the things that don’t show up in the price tag.
| Provider / tier | Monthly | Quota | €/min | Hosting | Source | Green |
|---|---|---|---|---|---|---|
| Otter Pro | $17 | 1,200 min | $0.014 | USA (CLOUD) | Closed | — |
| Otter Business | $30 | 6,000 min | $0.005 | USA (CLOUD) | Closed | — |
| Rev AI | PAYG | per-min | $0.25 | USA | Closed | — |
| Rev Human | PAYG | per-min | $1.50 | USA | Closed | — |
| Descript Hobby | $16 | 600 min | $0.027 | USA | Closed | — |
| Descript Creator | $24 | 1,800 min | $0.013 | USA | Closed | — |
| Trint Starter | $52 | file-capped | n/a* | USA / EU | Closed | — |
| Sonix | PAYG | per-hour | $0.167 | USA | Closed | — |
| Happy Scribe AI | PAYG | per-hour | €0.28 | EU (ES/IE) | Closed | — |
| Whisperit | undisclosed | — | — | CH (VÜPF risk) | Closed | — |
| Swiss Transcript | ~€20 | unclear | ~€0.07 | CH (VÜPF risk) | Closed | — |
| OpenAI Whisper API | PAYG | per-min | $0.006 | USA | Model open, API closed | — |
| SoTranscribe Self-Host | FREE | unlimited | €0 | Yours | Open | If your power is |
| SoTranscribe Hosted Light | €12 | 300 min | €0.040 | Austria | Open | ✓ 100% Ökostrom |
| SoTranscribe Hosted Plus | €25 | 1,500 min | €0.017 | Austria | Open | ✓ |
| SoTranscribe Patient Green | €8 | 300 min | €0.027 | Austria | Open | ✓✓ sun + wind only |
* Trint: file-capped — limited number of files per month (currently ~7 on Starter), so per-minute math depends on file length. n/a means we couldn’t fairly normalise.
Distilled and quantized Whisper variants run on much smaller hardware than you might think — even 8 GB Apple Silicon laptops or anything with an RTX 3060-class GPU is enough to self-host.
Honest take: on raw cost-per-minute we’re competitive, not radically cheaper than US incumbents. The differentiator is everything else in the table — Austrian jurisdiction, open source, 100% renewable. Pricing is also still in shaping — final hosted-tier numbers will be set transparently via Open Collective Europe before Beta launches.
Roadmap
Phase 1: ship a sovereign transcription pipeline that works.
The 10-month project (September 2026 – June 2027) is focused: open-source Whisper Large v3 + pyannote pipeline, Austrian hosting at Anexia Klagenfurt, four pricing tiers, EvoBioSys License. First paying beta customers by mid-2027.
Beyond Phase 1 we have research interests around confidential computing, renewable-energy-aware scheduling, and deeper hardware-sovereignty options — but we’re not making promises today on items we haven’t shipped. The honest framing: get the core pipeline production-ready first, then earn the right to talk about what comes next.
A note on “end-to-end encryption” in hosted transcription
Some competitors market “end-to-end encryption” for hosted transcription. We want to be honest about what that actually means. Audio in transit is encrypted (TLS). Audio at rest is encrypted on disk. But for the AI model to transcribe the audio, the audio must be decrypted into memory on the server — otherwise no model can compute on it. At that moment the plaintext is observable to anyone with privileged access to the running process: their sysadmins, anyone who compromises their server, or any government order that reaches the running memory.
This is not a vendor-specific problem. It’s structural for any hosted transcription service that uses standard hardware. The only architecture that fully solves it today is self-host — your audio never leaves your machine. We’re flagging this because we’d rather under-promise and let architecture do the work, than overclaim and hope nobody asks.
Who’s behind it
Built by SoTranscribe GmbH (in Gründung).
SoTranscribe is the first product spun out of idea2.life, an incubator within the EvoBioSys network. Co-founded by Katherine Colquitt (Co-Founder & Co-Projektleiterin; ex-Wall Street financial communications, VP Financial Services at Edelman Smithfield, leading clients including Citi, UBS, Schwab) and Jakob Possert-Bienzle (Co-Founder & Co-Projektleiter, Technical Lead). Vienna-based, EU/AT-jurisdiction by design.
Common questions
Plain answers.
Does anyone listen to my recordings?
No. The whole point is that nobody hears the audio except whoever you choose to share it with. The transcription is done by software, on hardware that’s either yours or ours — never on a service that re-uses your data.
How accurate is it?
Very. It uses the same family of speech-recognition models that the major services use, just running it without sending the audio elsewhere. For most clean recordings, accuracy is comparable to what you’d get from a paid cloud transcription service.
Why Austria?
Austria sits inside the EU and GDPR, but its surveillance law (§ 134 / 135a StPO) requires an individual court order before any provider can be compelled to release content. The Austrian Verfassungsgerichtshof struck down general data retention back in 2014. Stable, well-tested, court-supervised — the boring kind of jurisdiction you actually want for crown-jewel data.
Why not Switzerland?
A year ago we’d have hosted there too. In 2026 the picture has changed. The Swiss Federal Council’s pending VÜPF revision (Article 50a) would mandate provider-side decryption — a structural backdoor — and triggers data-retention plus government-ID identification at just 5,000 users, all via ordinance, bypassing parliament. Proton, the Swiss privacy flagship, is publicly moving infrastructure out of Switzerland into the EU. If their threat model says leave, ours says don’t arrive. We build directly in Austria.
What about privacy laws?
We’re built around GDPR. We don’t use any US-based company in the data path — not for payments, not for analytics, not for fonts. That’s a deliberate choice, not just a feature bullet.
What languages does it support?
About 100, with the strongest results in English, German, French, Spanish, Italian, Portuguese, Dutch, and the other major European languages. Detection is automatic.
Do I need to be technical to use it?
For the hosted version, no — upload an audio file, get text back. For self-host, yes — you’ll need to be comfortable running things on your own machine. The hosted version exists for everyone who’d rather just have it work.
Can I export my transcripts?
Always. Standard formats (Markdown + JSON). No lock-in. If you ever leave, you take everything.
When will it be live?
Open-source pipeline production-ready by mid-2027, with first paying beta customers. The 10-month build runs September 2026 – June 2027. Self-host code lands on GitHub during the build; hosted-tier waitlist opens ahead of beta.
Want the technical write-up? CLOUD Act reasoning, named providers, the full pipeline diagram, hardware research, lineage and ethos —
Talk to us
Mail is the front door.
- Mail — connect@sotranscribe.com (we read everything)
- Source — github.com/evobiosys/sotranscribe
- Matrix & Mastodon — coming soon