Lokas: Record and transcribe your meetings in complete privacy

3 days ago 3

“But your thing doesn’t work properly! It’s missing loads of features!”

Yes, you’re right 😀

That’s because Lokas isn’t (yet) an ‘official’ application. It’s not yet a ‘beta’ version (test version before release), or even an ‘alpha’ version (first draft). It’s a Proof of Concept (PoC).


What is a PoC ?

A PoC (‘Proof of Concept’, see previous FAQ), unlike standard software or applications, is not intended to be the first stone on the (long) road to a 1.0 version of this software.

It is a version that is deliberately developed very (but really very) quickly, in order to demonstrate a concept. Here, for Lokas: to demonstrate that Framasoft would have the capacity and skills to develop an application for recording and transcribing conversations, in complete confidentiality.

Think of a PoC as a showroom house, before you build a building: you can see how the house is organised, how the rooms are arranged, you can even walk around in it. However, it is NOT the house as it will be delivered. For example, furniture, electrical sockets and view are not what you will end up with. The showroom house is there to help you envision, BUT it is NOT the house that will be yours in the end.

As a result, you can come and tell us ‘I wouldn’t have put the sofa here’ or ‘The socket under the hood in the kitchen doesn’t work’ or ‘The accessibility of the bathroom leaves something to be desired’, and well… you’re right!

But above all, this means that, no, we’re not going to make this socket work, move the sofa or redo the bathroom. This is a showroom house, done in a hurry (read the Lokas story), and for the time being, there’s nothing to say that we’ll deliver such a house.


What are the limitations of Lokas?

At the moment, as this is a proof of concept, there are a number of limitations:

  • You are limited to 5 transcriptions per day
  • Transcriptions are limited to audio files of a maximum of one hour (60 minutes).
  • Lokas has an expiry date: if, by 1 May 2025, Framasoft has not raised enough money to get the project off the ground, we will shut down the transcript servers. (Note that the application is free and opensource, and its code accessible to anyone. So there’s nothing to stop new players extending the life of Lokas by providing machines to do so).
  • This is a PoC, and Framasoft reserves the right to modify these constraints or add new ones at any time.

Why use Lokas rather than other equivalent applications?

There are a number of applications for ‘Speech to Text’, also known as ‘audio transcription’, i.e. the conversion of vocal content into written text.

However, Lokas stands out in a number of ways:

  • Lokas is free software. Its source code is open. It can be duplicated, improved, re-shared, etc. by anyone who wants to and has the skills to do so;
  • Lokas support ‘diarization’. In other words, the application automatically tries to identify ‘who is speaking’ and ‘when’. For the moment, this is being done with varying degrees of success. Lokas is fairly effective at differentiating between a high-pitched voice and a low-pitched voice in a quiet environment, for example, but will probably fail with two voices with similar timbres in a noisy environment.
  • Lokas produces statistics per speaker. This was even its original aim. For example, to be able to easily objectify that men’s words are statistically often (largely) predominant in a meeting.
  • Lokas treats your data as completely confidential. So, unlike some applications that use their users⋅ices data (sounds and texts) to enrich the data models of the artificial intelligences of digital giants, your data is processed on Framasoft servers, and is not used for any purpose other than transcription, then automatically deleted from our servers.

When you say ‘in complete confidentiality’, what do you mean?

Framasoft, the publisher of Lokas, is an association under the French law of 1901 (nonprofits), and has been active for many years in the field of digital privacy. Our not-for-profit status and our ethical commitments mean that not only do we have no economic interest in monetising your data, but we also campaign politically against such practices.

In concrete terms, with Lokas :

  • You record your meeting: the audio file is stored on your phone;
  • You finalise your recording: a copy of this audio file is sent to our servers;
  • This file is put in a queue for transcription: how long it is kept depends on this queue;
  • The file is then transcribed into text: the text is sent back to your Lokas application;
  • The audio and text files are then deleted from our servers;
  • The text file is then stored on your phone;
  • You can delete the audio and text files on your phone at any time: Framasoft does not process these files.

In other words, your audio files only ‘leave’ your phone while they are being transcribed on our servers. The text file is then sent back to you, and we immediately delete these files from our servers.


Is Lokas RGPD compliant?

From a legal point of view: it’s a work in progress.

In practical terms, we are making progress to comply with the form imposed by the regulations, but the spirit of the law is being respected.

You can refer to our dedicated site rgpd.framasoft.org for generic information (logging and statistical monitoring), or for our thematic sheets (recipients, security, subcontractors, etc).

For the rest, the main data collected is the audio file, which is kept no longer than the end of the transcription.

For all requests for access, modification or deletion, please contact https://contact.framasoft.org.


What technologies does Lokas use?

The Lokas website uses free WordPress software.

The Lokas application uses code developed in-house under an open licence: https: //framagit.org/framasoft/lokas/lokas-app

The code for the Lokas server is also produced in-house under an open licence: https: //framagit.org/framasoft/lokas/server

On the transcription side, the ‘large-v2’ version of OpenAI Whisper (or Faster Whisper) is used, in conjunction with https://github.com/MahmoudAshraf97/whisper-diarization for diarisation, as well as NeMo and MarbleNet, also under open licences.


How do I report a bug or make a contribution?

Until early 2025, Lokas is really a proof of concept: we don’t plan to add any development time to it until we have the financial means to do so (based on your donations!).

However, you can ask your questions on our forum, or our contact form, without any guarantee of an answer.

If you’re a developer, you can also contribute directly to the code (application or server), but again, there’s no guarantee that your feedback will be taken into account.

Read Entire Article