AI drafts the record. A person approves it.
Audio and video are transcribed, with speakers attributed, on your own machines, even with no network at all. Your reviewer then corrects the draft in an editor built for long, exact work and approves the record. Nothing counts until a person signs it off.
From recording to approved record
The machine does the typing. A person decides what the record says. Nothing counts until they approve it.
Record or upload
Capture on desktop or phone, or bring the audio and video files you already have.
AI draft
Speech to text with speakers attributed, on your machine, on your servers or in our cloud, in the EU or the US.
Human review
Your reviewer plays the audio, fixes words, reassigns speakers, merges and splits turns.
Approve
A person signs the transcript off. Until then it stays a draft.
Translate and export
Translation memory and glossaries, then Word, subtitles or plain text.
An editor built for the record
Long, exact correction sessions, mostly from the keyboard. Audio and video follow the cursor, so checking a word takes a second.
I saw it on the fifteenth of March,
in the afternoon.
- Line length
- 35 / 37characters
- Lines
- 2 / 2
- Reading speed
- 12 / 15per second
- On screen
- 4.2 s
Illustrative · not a real recording
- Play or pause Space
- Replay the current line Shift + Space
- Speaker 1–9 for this line 1 – 9
- Next unreviewed line N
- Mark reviewed and move on Ctrl + Enter
- Split or merge a turn Ctrl + B / M
- Redact the selected words Ctrl + Shift + A
Three ways to run it
On a single machine, on your own servers or in our cloud. The same editor and the same results in each.
Air-gapped workstation
The desktop app with every model on the machine. It transcribes, separates speakers, translates and summarises with the network unplugged.
- Audio leaves the device
- No, the app needs no network
- Network needed
- No
- Best for
- Privileged and highly sensitive work
On your servers
Installed in your data centre or private network, so teams can share transcripts without them leaving your network.
- Audio leaves your network
- No, it stays on your servers
- Team sharing
- Inside your perimeter
- Best for
- Courts, firms, hospitals
Darka cloud
EU and US zones run on separate infrastructure, and recordings processed in one zone are not transferred to the other; your contract sets which one you use. Work you mark as sensitive runs only on speech and language models we host ourselves. For teams that need speed more than isolation. Cloud processing can be blocked for an organisation at any time.
- Processing region
- EU or US, set in your contract
- Zones
- EU and US, kept apart
- Best for
- Everyday, lower-risk work
What it does
From the recording to the document or the subtitles, in one place. People stay in the loop: every word can be checked against the recording.
Transcription
Speech to text from audio or video, in English, Polish and many other languages, with a timestamp on every line.
Device · server · cloudSpeaker attribution
Who said what, separated automatically. Rename a speaker once for the whole recording.
Device · server · cloudLine-by-line review
Every line is marked reviewed by a person, and the editor takes you straight to the next one that isn't.
Every deploymentApproval
The transcript stays a draft until a person approves it (labelled Certify in the app).
Every deploymentSubtitles
Cut from the reviewed transcript to streaming, broadcast, children's or vertical-video rules, with line length and reading speed checked on every cue. Existing SRT and WebVTT files can be imported and translated.
Every deploymentRedaction
Select words in the transcript and replace them with a label or a black bar before it leaves the team (labelled Anonymise in the app).
Every deploymentTranslation
Computer-assisted translation with translation memory and glossaries; Word, ODT and PDF in and out. Translations are drafts for a translator to approve.
Device · server · cloudExport
Word, SRT or WebVTT subtitles, or plain text, from the desktop apps for macOS, Windows and Linux, the Android app or the browser.
Every deploymentControls you can show an auditor
Confidentiality isn't only a promise in a policy. It is visible on every recording and enforced on the server, not only in the interface.
Cloud off, enforced
An organisation can have cloud processing blocked entirely. The server refuses it, not just the interface, and every change to the setting is logged.
Location on every item
Each recording shows where it lives, so nobody has to guess.
Honest about hardware
At setup it measures the machine and says what it can run and how fast, before a hearing, not during one.
Listed models
You get the list of models shipped and their licences, ready for your security review.
Built for sensitive recordings
Darka engineers install it on site, connect it to your case or records system, and keep it running.
Courts and tribunals
Hearing records transcribed on court hardware, with speakers named and every line timestamped.
Law firms
Depositions, client calls and interviews, kept on infrastructure you control.
Healthcare
Consultations turned into visit notes without patient audio leaving the clinic.
Police and public bodies
Interviews and recordings processed on isolated machines, under your own security rules.
Important information
Read this before you rely on a transcript. The licence and the data processing agreement govern each deployment.
- AI output is a draft. Every transcript, translation and summary produced by AI can contain errors and remains a draft until a person reviews and approves it. Rely only on reviewer-approved transcripts, and in proceedings only where the court or body accepts them under its own rules.
- Approval is your reviewer's sign-off. Approving a transcript (labelled Certify in the app) records that your reviewer checked it. It does not make Darka or the software a certified court reporter or sworn translator, and it is not an official record unless the court or body adopts it under its own rules.
- Recording is your responsibility. The customer must have a lawful basis, and any consent or notice the law requires, for every recording it processes.
- Not a medical device. Visit notes and summaries are documentation aids. The software does not diagnose, recommend treatment or replace clinical judgement.
- Performance depends on hardware. What runs offline, and how fast, depends on the machine. Setup measures it and shows the result before use.
- Where data is processed. In air-gapped and on-premises deployments, audio stays on the customer's hardware. In the Darka cloud it is processed in the zone set in the contract, the European Union or the United States; the two zones are kept apart. Subprocessors and transfer safeguards are listed in the data processing agreement. Darka does not hold security clearances and does not process classified information.
- Licensed technology. The service is delivered by Darka Software under licence. Third-party models are used under their own licences, listed in the third-party notices.
- Examples are illustrative. Screens, speakers and timestamps shown on this page are examples, not real recordings. Sectors named on this page describe who the product is built for, not existing clients.