How HoloTalk works

HoloTalk turns what a speaker says into live captions and translations that people in the room read on their own phones or on the big screen. It runs in a web browser: nobody installs an app, and nobody needs a headset.

On this page

Start a session

The host opens HoloTalk on a laptop or phone near the speaker and creates a session: a title, the language or languages the speakers will use, the caption languages to offer, and the most attendees to admit. The device's own microphone can be used, or an external microphone or the venue's sound feed selected as the device's microphone. Test the connection at the venue beforehand.

Speakers can use up to 3 languages in one session, such as English, Mandarin and Cantonese, and the host offers up to 10 translated caption languages at once, chosen from 45. The caption languages can be changed at any time, also during the session.

Add names and terms

Automatic captions can get names of people, places, organisations and awards wrong. The host can add them to the session's glossary, or import the programme or briefing documents, or paste text (PDFs with text, Word files in .docx format, text or Markdown files, up to 10 MB each), and choose from the names and terms HoloTalk proposes; only the text and the files' names are sent, never the files, and the names are not kept. The glossary is used when speech is recognised and when it is translated.

A session has a QR code and a join code of 8 characters. Attendees scan the QR code with their phone's camera, or type the code on the HoloTalk website, and choose their language. They need no account.

A private session has no QR code to share: the host invites people by email, up to 300 addresses, and each invitation has its own link. The join code alone lets nobody in.

Follow along live

While the speaker talks, each attendee reads the captions in the language they chose, with the original words under them if they want. Captions in the spoken language appear as the words are spoken, and each translation follows as soon as its sentence is complete. When later speech shows that a recent caption was misheard, it is corrected in every language.

A projector or big screen shows the latest captions in large type, in up to 3 languages at once; for a public session it also shows the QR code, so that people who arrive late can join.

On a computer, the speaker's console can also float the captions in a small window that stays above other windows, including slides played in a window; attendees can float them too, on a computer or a phone. For a live stream, the console gives a link to a transparent caption overlay for streaming software such as OBS or vMix.

If the speaker's connection drops, the console keeps up to 5 minutes of audio and sends it when the connection returns. Audio from a longer outage is saved on the speaker's device and sent after the session to be added to the transcript. Attendees reconnect by themselves.

After the session

When the host ends the session, HoloTalk writes a structured summary: an overview, key points, decisions, action items and open questions. If the host allows it, attendees can ask for the summary by email; once they confirm their address, they receive it in the language they chose.

The host, and attendees through the link in their email, can open the session's page to read the summary and the full transcript. Anyone with access can ask follow-up questions about what was said, paid for with their own credits; answers cite the parts of the transcript they rely on. The host can download the summary, the transcript and subtitles in the stored languages.

Uploading a recording instead

A host can also upload the recording of a meeting, audio or video, to get its transcript, translations and summary, with speaker labels if they ask for them. The page shows the most it can cost before anything is uploaded. The browser uploads only the sound, with the file's name; the picture of a video stays on the host's device.

What it costs

Sessions are paid for with credits, charged by the minute while captions run, at a rate set by the session's attendee limit and the number of translated languages being read. At an attendee limit of 50, an hour uses 780 credits with one translated language and 1,380 credits with 3 translated languages; smaller limits cost less and larger ones more. Paused time is not charged. The pricing page shows the rates by attendee limit, and the credit packs and plans.

The summary costs 14 credits for a session of up to an hour, 12 more for each further hour, and 4 for each language it is translated into. A follow-up answer costs 4 credits. An uploaded recording costs 2 credits a minute of audio, or 7 with speaker labels, plus its summary.

New accounts that verify their email address and complete registration can receive a signup bonus, currently enough for a session of 10 minutes in 3 languages; the conditions are in the Terms of Service.

Before a large event

Open the join link on the venue's Wi-Fi beforehand.
Put the speaker's device on a wired or reliable connection, with its charger plugged in.
Set the attendee limit to the number of people you expect.
Add the speakers' and organisations' names to the glossary.
Tell the audience that the session is captioned and translated by AI: the Acceptable Use Policy has a notice you can show or read.
Where accuracy is critical, such as medical, legal or safety information, also book a qualified interpreter.