Local meeting transcription in Barcelona, no cloud
Meetings, consultations and voice notes turned into text, with who said what, a summary and the tasks. All inside your server.
We install 100% local voice transcription for law firms, clinics and teams, ready in 2 to 3 weeks: meetings, consultations and voice notes become text showing who is speaking, with a summary and tasks, without the audio leaving your server. It works in Spanish, Catalan and English, even when they are mixed in the same meeting.
Law firms that record client meetings and don’t want to upload the audio to the cloud
Clinics and practices that want to spend less time writing notes
Management teams and committees that need reliable minutes
Public bodies and educational organisations with sessions in Catalan and Spanish
People who work in the field or on the move and prefer dictating to typing
What is included
Local Whisper transcription engine, configured for Spanish, Catalan and English
Speaker diarisation with WhisperX: separates and labels each person who speaks
Summary, decisions and tasks generated by a local model
Minutes or note templates for your type of meeting or consultation
Inbox folder or internal page for uploading audio and video files
Optional: voice notes from your phone via Telegram, Discord or a web app
Automatic deletion of the original audio according to the policy you choose
What you get
Transcription service running on your server
Minutes and summary templates adapted to your work
Quality report using your own test recordings
User guide and template notice to inform people about the recording
The problem: audio is the most sensitive data of all
In a client meeting, a consultation or a committee, people say things they would never put in an email. Many transcription services and meeting assistants send that audio to servers outside your control, sometimes outside the EU. Under the GDPR, a recording with identifiable voices is personal data; if it contains health data, it is a special category; and if you work under professional secrecy, it shouldn’t pass through anyone else’s hands. What’s more, if you generate summaries with AI and send them to clients, it is wise to make clear that they were prepared by an AI, in line with the EU AI Act (Art. 50).
Local transcription turns voice into text inside your office. The audio doesn’t travel, there is no per-minute fee and you can delete it as soon as you have the text.
How we do it, step by step
We collect real test recordings, with permission, from your meetings or consultations.
We choose the model and settings that work best with your language, your accents and your rooms.
We install the service on your AI server or on a dedicated machine with a GPU.
We design the templates: meeting minutes, consultation note, client summary or task list.
We connect the input: a folder, an internal web page or voice notes from your phone.
We measure and adjust with your recordings, and leave you a template notice to inform people about the recording.
The technology, in plain words
Whisper: the most widely used open speech recognition model. It runs on a GPU and transcribes an hour of audio in a few minutes.
WhisperX with diarisation: adds precise timestamps and separates the participants (“speaker 1”, “speaker 2”), whom you can then rename.
A local language model, served with vLLM or Ollama, reads the transcript and drafts the summary, decisions and tasks.
Catalan: Whisper works well out of the box, even when a meeting in Barcelona mixes Catalan and Spanish. If your case calls for it, we also consider the open models from the AINA project.
Everything can run on the same server as your local AI, and the transcripts can feed your private RAG search so you can later ask “what did we agree with this client in March?”.
When we DON’T recommend it
If your meetings aren’t confidential and you already pay for a video-conferencing suite with built-in transcription and a data processing agreement, use it. If you only record a few hours a month, installing anything may not be worth it. And if you need a transcript with legal standing, for example for court proceedings, AI helps a great deal, but human review will always be required.
We use it every day
In our Barcelona lab we transcribe meetings with diarisation entirely locally, from start to finish, without the audio ever touching the cloud. And on a family estate in the Penedès, the family logs irrigation, treatments and expenses with voice notes in Catalan, which an agent turns into text and stores in its field logbook. If it works in the middle of a vineyard, it will work in your meeting room.
A field agent that speaks Catalan, knows your farm and keeps the logbook for you.
Voice field logbook
Local weather and history
Pests and diseases from photos
Learn more →
Frequently asked questions
What hardware do I need for local transcription?
A machine with a GPU. If you already have local AI, it is installed on the same server; if not, on a dedicated machine we help you choose. With a GPU, Whisper transcribes an hour of audio in a few minutes, and written minutes, calendar integration or voice notes from your phone are added on top.
Does it work in Catalan?
Yes. Whisper transcribes Catalan, Spanish and English well, even when they are mixed in the same meeting, which is very common in Barcelona. We check the quality with your own recordings before handover.
Is it as accurate as a cloud service?
With good audio, accuracy is comparable. What matters most is the microphone and the noise in the room, not where the processing happens; that’s why we also advise you on recording equipment.
Can I use it in medical consultations?
It can help prepare the notes, but the clinician must always review what goes into the medical record and inform the patient about the recording. Because the audio never leaves your practice, the privacy risk is much lower.
Free assessment
We install 100% local voice transcription for law firms, clinics and teams, ready in 2 to 3 weeks: meetings, consultations and voice notes become text showing who is speaking, with a summary and tasks, without the audio leaving your server. It works in Spanish, Catalan and English, even when they are mixed in the same meeting.