On this page

Live Captions and Translation API

People speak their own language. Everyone else reads captions or hears the speaker in theirs, a few seconds later.

Meet does the speech to text and the translation for you. There is nothing to connect and no extra key. You turn it on for a room, and each person picks what they want to read or hear.

What each part does#

Captions. Words appear as people speak. Partial text shows first and is replaced by the final line.
Spoken translation. A listener chooses a language and hears the speaker in that language. Each listener and language is counted, see Pricing.
Transcript. Final caption lines are stored with the session, with their translations, and you can download them later.

1. Turn it on for the room#

bash

curl "https://meet.respiresoft.com/api/v1/rooms" \
  -H "Authorization: Bearer $MEET_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "town-hall-1", "privacy": "private", "properties": {
        "enable_captions": true, "enable_translation": true, "languages": ["es", "fr", "ar"] }}'

languages lists the translation languages for the room, at most 5, and at least one when enable_translation is true. Setting both enable_captions and enable_translation to false switches speech off for the room.

2. Say what language each person speaks#

Put lang in the meeting token so the speaker is understood in the right language from the first word:

bash

curl "https://meet.respiresoft.com/api/v1/rooms/town-hall-1/tokens" \
  -H "Authorization: Bearer $MEET_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "Ann", "role": "participant", "lang": "en", "perms": ["audio", "video", "chat", "translate"]}'

3. Use it from your page#

With the embed SDK:

await call.setCaptions(true);                       // starts `caption` events and shows captions
await call.setLanguage('es');                       // hear the call in Spanish
call.on('caption', ({ participantId, text, final, lang }) => {
  if (final) console.log(participantId, lang, text);
});

The call screen also has its own captions and language menu, so you can use them without writing code.

4. Get the transcript afterwards#

bash

curl "https://meet.respiresoft.com/api/v1/sessions/$SESSION_ID/transcript?format=vtt&lang=es" \
  -H "Authorization: Bearer $MEET_API_KEY"

format is json (default, paginated, with start_ms and end_ms from the start of the session), txt, vtt or srt. lang shows the translation where a line has one. A transcript.ready webhook tells you when a finished session has lines.

Good to know#

Captions come from live speech recognition, so names, noise and accents can cost accuracy. Give a transcript a read before you publish it.
Spoken translation is not instant. The listener hears each sentence a few seconds after the speaker, because the meaning has to take shape before it can be translated and spoken.
The call screen has 18 interface languages. Captions and translation languages are chosen separately, per room.