RO

All projectsCase study Our own product

LinguaLive.

The speaker talks and every listener reads the talk on their own phone, in their own language, sentence by sentence. From the same phone they can ask, speak, vote and say how it went. No app and no headset.

For
Seminars, courses, conferences and public meetings
What we did
Research, design and development
Status
Public demo, since November 2026
Address
socialit.ro/explore/lingua (new tab)
The LinguaLive home page: the headline “Every listener reads the talk in their own language”, the field for the session code and a caption translated from Romanian into English.
The home page. Screenshots come from the public page and from a demo started on it, on 11 and 12 November 2026.

How it works.

  1. The invite

    The organiser receives a link for a session of 10 to 90 minutes. They choose the spoken language and add up to 40 names and terms that must not be mangled.

  2. The code

    A six-character code or a QR code, on a slide or on the projection screen.

  3. Speaking

    Each sentence appears while it is spoken, then settles into its final form.

  4. A second read

    A few seconds later, an AI reviser mends the punctuation and the misheard names. A word it adds is marked.

  5. 15 languages

    Each listener picks their language and can change it at any time. The names on the list are not translated.

  6. At the end

    The transcript can be downloaded, as heard or in clean paragraphs. After 15 minutes it is deleted from the server.

The “Set up your time on air” screen: the session title, the spoken language, the speaker's name, the list of names and the audio source.
Setting up the session, in a demo started on the site
The speaker console: the QR code and the session code, the “On air” button, the current sentence and the one before it, in Romanian.
The speaker console, in the same demo
The projection screen: two large sentences in Romanian, the QR code and the join code.
The projection screen
The “Where it fits” section: five kinds of event, with the seminar open and the chain microphone, laptop, phones.
Where it fits
The “What you get” section: eight things every session has, from the live text to the deletion of the transcript.
What every session has
The “Sound for your room” page: five kinds of room, with the hall with a sound system open and what to ask from the mixing desk.
The sound guide, by kind of room

On the listener's phone.

The listener scans the QR code or types the six-character code. They install nothing and create no account.

They pick one of the 15 languages and can change it at any time. The current sentence sits at the top, large, and the earlier ones below.

Three buttons show the original, enlarge the letters and save the text. Words added by the reviser appear with a dotted underline, with an explanation.

The listener's page on a phone: the current sentence in English, translated from Romanian, and the sentence before it.
Romanian spoken, English read, in the same demo
The listener's page on a phone, with a word with a dotted underline and the note “Dotted words were added in revision”.
What a word added by the reviser looks like: a trial session, with sample sentences and translations

The room answers back.

On the same code, people in the room can ask, speak, vote and say how it went. The moderator decides when.

Phone mock-up: a question written in Romanian and, underneath, the same question in English, on the moderator's screen.
Written questions. The listener writes in the language they read in. The moderator sees the question translated and reads it out.
Phone mock-up: the message “You are live in the room. Speak now.” and, on the moderator's screen, the name of the person speaking.
The floor, from the phone. When the moderator gives it, the phone becomes the microphone and the voice is heard on the room's loudspeakers.
Phone mock-up: a poll in Romanian with three answers and percentages and, underneath, the question in English, as the moderator wrote it.
Polls. The moderator writes a question with two to five answers. Everyone sees it in their own language and picks one.
Phone mock-up: the rating from 1 to 5 and a comment and, underneath, the average for the organiser.
Feedback. At the end, each phone asks for a rating from 1 to 5 and a few words, with no name.

The four phones are the mock-ups from the public page, with sample data.

Technical decisions.

  • Sessions that disappear

    An invite starts a single session, of a fixed length. Fifteen minutes after the end, the session, the transcript and the translations are deleted from the server.

  • Whole sentences

    A sentence is closed only at a real pause after a full stop or when it grows too long, so that a breath does not split it in two.

  • A reviser in two steps

    The text appears at once. A few seconds later, a Claude model corrects it, with the last three sentences as context. The words it adds are marked.

  • The sound does not pass through our server

    The speaker's microphone goes from the browser straight to the speech-to-text service. In the public demo, the browser only receives a short-lived key.

  • A live voice, not a recording

    When a listener is given the floor, the sound goes straight from the phone to the speaker's laptop. It is not recorded.

  • No account and no database

    Everything runs in the browser. The server is plain PHP: a session is a set of files that get deleted.

What stays private.

  • The sound is not kept

    It goes to Deepgram, which turns it into text. We do not record it and we do not store it.

  • The text is deleted

    It stays on the server for as long as the session runs and 15 minutes more. In the public demo, two minutes.

  • Listeners have no account

    They give no name and no email. We only count how many browsers joined a session.

  • Only the moderator sees the questions

    The name is optional. Questions are deleted with the session.

  • Polls and feedback with no name

    A vote is stored next to a random number, not next to a name. Feedback is anonymous.

  • What reaches others

    Deepgram receives the sound, Google Cloud Translation the sentences to translate, and Anthropic the sentences to correct. These services may run outside the European Union.

Status.

LinguaLive has had a public demo since November 2026: three minutes of speaking, up to five phones reading along, and everything is deleted two minutes after the end. A real event starts from an invite link, which we send on request. It has not yet been used in a room, with an audience. The interface is in English.

What it does not do: it does not speak, listeners read. The translation is automatic and gets things wrong, most often the names it was not given. It needs the internet and a microphone close to the speaker.

Write to us.

Do you have a procedure that would run more easily in an app? Tell us in a few lines what you want to solve. You do not need a prepared document. We will then set up a first conversation.

contact@socialit.ro
or call +40 752 091 605