Ditara · Offline speech-to-text
An offline speech-to-text system that turns dictations into editable text drafts — automatically, on your own machines, with no cloud and no AI chat.
Inside your building
Recorder
the dictaphone you already use
Shared folder
where recordings already land
Text draft
waiting before anyone opens it
No cloud
Speak instead of type, and the audio never leaves your building.
The problem
Professionals dictate on recorders. Someone else listens and types it up, slowly, between all their other tasks.
Health data, legal matters, clients' personal details. Sending them to a third-party AI service isn't acceptable, legally or to the people involved.
Dictaphones save in proprietary formats that common audio tools decode as noise.
Whatever the solution, the same recorders, the same folders and the same people had to stay in place.
The solution
Before
After
Speak, don't write — without talking to an AI. There's no chatbot, no prompt, no account and no cloud. You talk to a recorder as you always did, and a text draft appears in a folder.
What it does
100% offline
The speech model runs on your own computer. Audio and text never leave the building, and it works with the network cable unplugged.
No AI chat, no accounts
Nobody types prompts or talks to a bot. The AI is a component inside, not an interface.
Your recordings are never touched
It never moves, renames or edits the originals. Drafts go to a separate folder.
A person always signs off
Output is always a draft to check. Corrections are saved separately, and the original draft is kept for comparison.
Reads dictaphone formats
Compatible with .dss and .ds2 recordings from Olympus, Philips and Grundig dictaphones, plus .wav and .mp3.
No more noise halfway through
Popular tools turn many of these recordings into noise partway in. This decoder reads them cleanly, including paused, resumed and edited recordings.
Encrypted recordings supported
It asks for the password when a recording needs one.
Drafts appear on their own
Point it at the folder recordings already land in. Every new dictation becomes a text draft in the background, once the file has finished copying.
Nothing is done twice
It reads the dictation system's own records to skip work already transcribed, and recognises the same recording even if it's renamed or uploaded twice.
Organised output
Drafts are filed per author, each with a small record of date, source and software version for traceability.
Starts itself, stays out of the way
It runs hidden in the background. A status line shows what it's doing, with Stop and Start buttons on the page.
Survives problems
It logs every step, runs only once per folder, and reports the last warning or error in plain language.
Speaks your punctuation
Spoken commands like "full stop", "comma" and "new paragraph" become punctuation, and pauses become paragraphs. The result reads like a report, not a transcript.
Learns your vocabulary
A short list of medical, legal or technical terms steers the model toward the right spelling.
Multilingual
Built for European Portuguese. The model handles dozens of languages.
A review page
Read, edit, copy and save drafts in the browser, in a window sized to sit beside the audio player. You can also drop files in directly.
A desktop app for one person
Pick files, press Transcribe, copy the text. For a single professional with no IT department.
The review page
For the IT department
Speak, don't write
Any business where people write reports from what they saw, heard or decided, and where that information is confidential.
Clinics & hospitals
Imaging and consultation reports, discharge notes.
Law firms & notaries
Case notes, memos, client meeting summaries.
Insurance & loss adjusters
Site-visit and claim reports.
Inspectors, surveyors, engineers
Inspection reports dictated on site.
Psychologists & therapists
Session notes that must stay private.
Veterinarians & dentists
Clinical notes between appointments.
HR & recruitment
Interview notes.
Researchers & journalists
Interview transcription, with sources protected.
Field sales & technicians
Visit and service reports dictated from the car.
Public administration
Minutes and records that can't go to foreign clouds.
If your team already records, or would rather talk than type, but can't send that audio to the cloud, this is the shape of the solution.
How we'd start
A short conversation
You tell me how reports get written today: who records, where the files go, who types them up. Rule number one: nobody has to change how they work.
A test on your recordings
We run a handful of your own recordings through Ditara, on your premises, so you judge the drafts for yourself.
Fitted to your work
Your folders, your authors, your vocabulary and your language, set up so the drafts read like your reports.
Installed with your IT
On one PC or an on-site server. Your IT team runs the self-check, and nothing connects to the internet.
A safe trial run
Ditara runs beside your current process first. You compare its drafts with your final reports, and switch over only when you're happy.
Your turn
Confidential recordings, notes or forms that someone types up again? Let's talk about automating it, on your own machines.
Start a conversation