Jitsi caption languages and live translation setup

Who this is for

Use this after captions work. Start with the Vosk guide or bridge guide for recognition, or our transcription service for setup help. [S17][S18]

How it works

Recognition converts speech into text. Translation converts that text into another language. Jitsi initializes the speaker’s recognition language from the application language when transcription.useAppLanguage is true. Otherwise it uses transcription.preferredLanguage, with en-US as the fallback for an unsupported selection. [S3]

Each reader selects a caption target under Translate to, or Original. The client publishes translation_language; the spoken language uses transcription_language. Jigasi translates final results into requested targets. Its Whisper adapter is an exception to clean separation: it prefers the participant’s translation target when constructing the recognition hint. [S3][S6][S17]

Language lists are UI choices, not installed models or access controls. Use full speech tags such as pl-PL for Jigasi Google recognition, and translator-supported target codes such as pl. [S3][S14][S17]

Before you start

Checked October 6, 2026: stable-11248, released September 14, remains Docker’s latest release. Steps target it and package Jigasi 1.1-415-g2750d54-1. Preserve authentication, recognition and existing overlays. Recreate outside active meetings. [S1][S4]

Frontend custom languages shipped through Meet PR #16984. Docker’s TRANSCRIPTION_CUSTOM_LANGUAGES change, PR #2234, merged September 22, after stable-11248. Its patch does not add Compose forwarding. Do not expect that variable in stock stable-11248. [S15]

Prepare a translation provider and supported source/target pairs. Google requires its API and billing; Microsoft requires a Translator resource. Confirm participant agreement before sending text to an external provider. [S10][S11]

Steps

Docker stable-11248: configure the client. Put these assignments in .env. This example starts recognition in English and offers three caption targets. [S2]
dotenv
```
ENABLE_TRANSCRIPTIONS=1
USE_APP_LANGUAGE=false
PREFERRED_LANGUAGE=en-US
TRANSLATION_LANGUAGES=['en','de','pl']
TRANSLATION_LANGUAGES_HEAD=['en']
JIGASI_TRANSCRIBER_ENABLE_TRANSLATION=1
```
USE_APP_LANGUAGE=true initializes speech from the interface language. Settings > General > Language updates the live language, but sends bare codes such as pl; released Google recognition rejects these. Use a full spoken tag before joining, or the API property command below. [S3][S16][S17]

Use your actual CONFIG; this example uses the handbook default. Add these optional custom speech entries only with matching backend models. [S3][S4]
Terminal
```
CONFIG="$HOME/.jitsi-meet-cfg"
mkdir -p "${CONFIG}/web"
cat >> "${CONFIG}/web/custom-config.js" <<'EOF'
config.transcription.translationEnabled = true;
config.transcription.customLanguages = {
    ...config.transcription.customLanguages,
    'hsb-DE': 'Upper Sorbian (Germany)',
    'dsb-DE': 'Lower Sorbian (Germany)'
};
EOF
sudo chown 1000:1000 "${CONFIG}/web/custom-config.js"
sudo chmod 644 "${CONFIG}/web/custom-config.js"
```
Preserve existing custom entries. Startup appends this file after generated config.js; rootless web reads it as UID 1000. [S4]
Docker stable-11248: choose one translator. Set its class in .env, preserving your speech service. [S5][S6]

Google: [S7][S10]
dotenv
```
JIGASI_TRANSCRIBER_CUSTOM_TRANSLATION_SERVICE=org.jitsi.jigasi.transcription.GoogleCloudTranslationService
GOOGLE_APPLICATION_CREDENTIALS=/run/jigasi/config/google-key.json
```
Enable Cloud Translation API and billing. Save private ADC JSON at the chosen filename ${CONFIG}/transcriber/google-key.json; startup copies it. Run sudo chown 1000:1000 "${CONFIG}/transcriber/google-key.json" and sudo chmod 600 "${CONFIG}/transcriber/google-key.json". Do not commit it. [S4][S7][S10]

Stock transcriber.yml does not forward GOOGLE_APPLICATION_CREDENTIALS. Create this guide-specific overlay, language.yml, and include it in subsequent Compose commands: [S4][S5]
YAML
```
services:
  transcriber:
    environment:
      - GOOGLE_APPLICATION_CREDENTIALS
```
Microsoft Translator: its Java implementation shipped through Jigasi PR #636. Docker PR #2251 closed without merging, so use the existing custom-service variable and properties. [S8]
dotenv
```
JIGASI_TRANSCRIBER_CUSTOM_TRANSLATION_SERVICE=org.jitsi.jigasi.transcription.BingTranslationService
```
Append these to ${CONFIG}/transcriber/custom-sip-communicator.properties, replacing placeholders with your resource’s key and region. Run sudo chown 1000:1000 "${CONFIG}/transcriber/custom-sip-communicator.properties" and sudo chmod 600 "${CONFIG}/transcriber/custom-sip-communicator.properties". [S4][S8][S11]
Properties
```
org.jitsi.jigasi.transcription.bing.subscription_key=REPLACE_WITH_TRANSLATOR_KEY
org.jitsi.jigasi.transcription.bing.subscription_region=REPLACE_WITH_RESOURCE_REGION
```
LibreTranslate: [S9]
dotenv
```
JIGASI_TRANSCRIBER_CUSTOM_TRANSLATION_SERVICE=org.jitsi.jigasi.transcription.LibreTranslateTranslationService
JIGASI_TRANSCRIBER_LIBRETRANSLATE_URL=http://libretranslate:5000/translate
```
This private language.yml adaptation publishes no host port. The base Compose file supplies the network; LibreTranslate loads the selected models. [S5][S12]
YAML
```
services:
  libretranslate:
    image: libretranslate/libretranslate:v1.9.6
    environment:
      LT_LOAD_ONLY: en,de,pl
    networks:
      - meet.jitsi
```
Jigasi sends an empty api_key. A key-required endpoint cannot work unchanged; use a private endpoint accepting keyless requests or modify the adapter. [S9]
Docker stable-11248: apply the selected setup. For Google or the LibreTranslate example, include language.yml; otherwise omit that extra file. Include your other existing overlays too. [S4][S5]
Terminal
```
docker compose -f docker-compose.yml -f transcriber.yml -f language.yml up -d
docker compose -f docker-compose.yml -f transcriber.yml -f language.yml up -d --force-recreate web transcriber
```
Reload and join a fresh meeting. [S3][S4]
Debian/Ubuntu packages, Meet 11248 and Jigasi 1.1-415: edit /etc/jitsi/meet/meet.example.com-config.js. Use the same frontend properties directly, without Docker environment variables: [S3][S13]
JavaScript
```
config.transcription = {
    ...config.transcription,
    enabled: true,
    translationEnabled: true,
    useAppLanguage: false,
    preferredLanguage: 'en-US',
    translationLanguages: ['en', 'de', 'pl'],
    translationLanguagesHead: ['en']
};
```
Add customLanguages inside that object only for a capable backend. In /etc/jitsi/jigasi/sip-communicator.properties, enable translation and select one class from step 2: [S6][S13]
Properties
```
org.jitsi.jigasi.transcription.ENABLE_TRANSLATION=true
org.jitsi.jigasi.transcription.translationService=org.jitsi.jigasi.transcription.GoogleCloudTranslationService
```
For Microsoft, add the same bing properties; for LibreTranslate, set org.jitsi.jigasi.transcription.libreTranslate.api_url to your reachable endpoint. For Google, place GOOGLE_APPLICATION_CREDENTIALS=/etc/jitsi/jigasi/google-key.json in /etc/jitsi/jigasi/config; the JSON filename is chosen here. Protect the saved JSON, then restart: [S7][S8][S9][S10][S13]
Terminal
```
sudo chown jigasi:jigasi /etc/jitsi/jigasi/google-key.json
sudo chmod 600 /etc/jitsi/jigasi/google-key.json
sudo systemctl restart jigasi
```
Other translators need only the restart command. [S13]
Bridge proxy source 6747f187, either installation: configure its own text translator. This Google example belongs in the proxy environment, with the key replaced locally: [S18]
dotenv
```
ENABLE_TEXT_TRANSLATION=true
TEXT_TRANSLATION_PROVIDERS_PRIORITY=google
TEXT_TRANSLATION_GOOGLE_API_KEY=REPLACE_WITH_GOOGLE_TRANSLATION_API_KEY
```
Docker: forward these three variables in the proxy’s environment list and recreate it. Packages: update the proxy’s dedicated environment file and recreate its container, following the bridge guide. Other text providers are OpenAI, Gemini and xAI. Translation processes finals, separately from translated audio. [S18]

Configuration reference

Client names below are inside config.transcription. Docker defaults apply to stock stable-11248; package defaults may differ from its generated values. [S2][S3]

Name, Docker mapping	Location	Default	Purpose
`CONFIG`	Docker `.env`	`~/.jitsi-meet-cfg`	Host configuration folder. [S4]
`enabled`, `ENABLE_TRANSCRIPTIONS`	Client, `.env`	`false`	Enable feature; Docker also sets `disableClosedCaptions` to its inverse. [S2]
`useAppLanguage`, `USE_APP_LANGUAGE`	Client, `.env`	`true`	Initialize speech from app language. [S2][S3]
`preferredLanguage`, `PREFERRED_LANGUAGE`	Client, `.env`	`en-US`	Initial speech when app-language mode is false. [S2][S3]
`translationEnabled`	Client custom config	`true`	Show translation selection UI. [S3]
`translationLanguages`, `TRANSLATION_LANGUAGES`	Client, `.env`	Docker `[]`	Caption target list. [S2]
`translationLanguagesHead`, `TRANSLATION_LANGUAGES_HEAD`	Client, `.env`	`['en']`	Targets listed first; stock Docker offers English. [S2][S3]
`customLanguages`, `TRANSCRIPTION_CUSTOM_LANGUAGES`	Client custom config	`{}`	Additional accepted speech tags; Docker shortcut unreleased. [S3][S15]
`autoTranscribeOnRecord`	Client custom config	`false`	Start transcription with recording. [S3]
`autoCaptionOnTranscribe`	Client custom config	`false`	Display captions when transcription starts. [S3]
`renderTranscriptDetails`, `diarize`	Client custom config	`false`	Transcript details and requested diarization; backend-dependent. [S3]
`disableStartForAll`, `DISABLE_START_FOR_ALL`	Client, `.env`	`false`	Historical nested option emitted by Docker. [S2]
`autoCaptionOnRecord`, `AUTO_CAPTION_ON_RECORD`	Client, `.env`	`false`	Historical nested option; not modern `autoTranscribeOnRecord`. [S2][S16]
`JIGASI_TRANSCRIBER_ENABLE_TRANSLATION`	`.env`	`0`	Maps to `org.jitsi.jigasi.transcription.ENABLE_TRANSLATION`, default false. [S5][S6]
`JIGASI_TRANSCRIBER_CUSTOM_TRANSLATION_SERVICE`	`.env`	Unset	Maps to `org.jitsi.jigasi.transcription.translationService`; fallback Google. [S5][S6]
`JIGASI_TRANSCRIBER_LIBRETRANSLATE_URL`	`.env`	Unset	Maps to `org.jitsi.jigasi.transcription.libreTranslate.api_url`; adapter default `http://libretranslate:5000/translate`. [S5][S9]
`org.jitsi.jigasi.transcription.bing.subscription_key`, `org.jitsi.jigasi.transcription.bing.subscription_region`	Jigasi properties	Empty	Microsoft credentials. [S8]
`GOOGLE_APPLICATION_CREDENTIALS`	Process environment	Docker `/run/jigasi/config/key.json`	Google ADC JSON path. [S4][S10]
`LT_LOAD_ONLY`	LibreTranslate environment	Unset	Load selected translation models. [S12]
`ENABLE_TEXT_TRANSLATION`	Bridge proxy environment	`false`	Enable caption text translation. [S18]
`TEXT_TRANSLATION_PROVIDERS_PRIORITY`	Bridge proxy environment	`openai,gemini,xai,google`	Provider selection order. [S18]
`TEXT_TRANSLATION_GOOGLE_API_KEY`	Bridge proxy environment	Unset	Google translation credential. [S18]
`transcription_language`, `translation_language`	Participant properties	Initialized speech tag, no target	Speech locale and caption target respectively. [S3][S16]

Common mistakes

Only one translation direction works. Issue #635 reports Meet 2.0.10888-1 and Jigasi 1.1.407-g1e3a7ac, with Google recognition and translation. Its exact reported error is: [S14]

Text

UnsupportedOperationException: pl is not a language supported
by the Google Cloud speech-to-text API

The target/STT mix-up is unconfirmed. Released Google recognition rejects bare pl; use pl-PL for speech and inspect participant properties. PR #640 changes recognition locale resolution and remains open. No complete workaround is confirmed; #635 is not fixed as of 2026-10-06. [S14][S17]

Runtime overrides do not update speech language. #16172 was auto-closed without a fix. Initial configOverwrite or URL overrides apply before conference creation; runtime overwriteConfig does not reapply speech language. These fresh-join URLs set English or Polish for that participant, not the whole room: [S3][S16]

Text

https://meet.example.com/language-test#config.transcription.useAppLanguage=false&config.transcription.preferredLanguage=%22en-US%22
https://meet.example.com/language-test#config.transcription.useAppLanguage=false&config.transcription.preferredLanguage=%22pl-PL%22

An embedded app can write the participant property directly. This is source-confirmed, not meeting-tested: [S16]

JavaScript

api.executeCommand('setParticipantProperties', {
    transcription_language: 'de-DE'
});

More choices, same model. Vosk supports loaded models, with per-language routing available through Jigasi. Whisper needs a multilingual model, not .en. Google’s Jigasi adapter has its own accepted-tag list. Bridge support depends on the provider and model; Deepgram’s default Nova-2 multilingual mode covers English and Spanish. [S17][S18]

Verify

For both installations, inspect the served configuration rather than only the host file: [S2][S4][S13]

Terminal

curl -fsS https://meet.example.com/config.js

Expect useAppLanguage: false,, preferredLanguage: 'en-US', and translationLanguages: ['en','de','pl'], for Docker. This proves delivery, not recognition. Refresh and bypass CDN caching. [S2][S4]

Test English and Polish speakers using the fresh-join URLs above. Require translated final captions in both directions, then check Original. This is your acceptance test; no successful meeting was observed for this guide. [S3][S6][S14][S16]

If it still fails

For Docker, run docker compose -f docker-compose.yml -f transcriber.yml logs --tail=200 transcriber, adding your overlays. Packages write /var/log/jitsi/jigasi.log; use sudo journalctl -u jigasi -n 200 --no-pager for service failures. Look for Cannot instantiate custom translation service, which causes fallback to Google, and the exact unsupported-language error above. [S6][S13][S14]

Check provider authentication, enabled APIs, target language support and network access separately from speech recognition. Use the existing transcription troubleshooting page when original captions also fail. [S7][S8][S9][S10][S11]

FAQ

Does changing the Jitsi interface language change captions?

With useAppLanguage=true, it initializes speech language from that interface language. The web settings also update spoken language during a conference; each reader’s translation target is a separate choice. [S3]

Can I add any language with customLanguages?

It extends accepted speech tags, not the interface-language dropdown or installed models. Use a capable backend. [S3][S17]

Can I translate without Google?

Yes. Released Jigasi includes Microsoft and LibreTranslate adapters. Bridge caption translation uses its own providers and configuration. [S8][S9][S18]

Is the one-way Polish translation issue fixed?

No confirmed released fix was found as of October 6, 2026. #635 and PR #640 remain open; testing full speech locale tags is diagnostic guidance, not a guaranteed solution. [S14]

Sources

[S1] Docker latest release, September 14, 2026, release note; stable package index, checked October 6, 2026, official repository.

[S2] Docker web settings, Compose forwarding, September 14, 2026, source code.

[S3] Meet configuration, language selection, settings update, caption picker, SDK language property, September 2026, source code.

[S4] Docker handbook, checked October 6, 2026, official doc; web startup, Jigasi startup, image credentials default, September 14, 2026, source code.

[S5] Transcriber template, transcriber Compose, September 14, 2026, source code.

[S6] Jigasi translator selection, final-result translation, September 11, 2026, source code.

[S7] Google translator, September 11, 2026, source code.

[S8] Microsoft translator, source code; Jigasi PR #636, May 8, 2026, official repository PR; Docker PR #2251, August 17, 2026, unmerged repository PR.

[S9] LibreTranslate adapter, September 11, 2026, source code.

[S10] Google translation setup, ADC lookup, checked October 6, 2026, official docs.

[S11] Microsoft Translator authentication, checked October 6, 2026, official doc.

[S12] LibreTranslate v1.9.6 Compose example, published image, defaults, May 2026, source code and official registry.

[S13] Jitsi package quickstart, official doc, checked October 6, 2026; Jigasi service, sample properties, September 11, 2026, source code.

[S14] One-way translation #635, April 20, 2026, community report; PR #640 and maintainer discussion, June 1, 2026, proposed source change and maintainer comments; statuses checked October 6, 2026.

[S15] Meet custom-language PR #16984, February 25, 2026; request #16856, closed March 26, 2026; Docker PR #2234, merged September 22, 2026, official repository changes and maintainer comments.

[S16] Runtime override #16172, June 2025, community report, auto-closed November 7, 2025; reducer, initialization, API, initial URL config, parser, whitelist, September 2026, source code.

[S17] Vosk models, Whisper models, official docs; Google validation, Vosk routing, Whisper hint, September 11, 2026, source code.

[S18] Proxy configuration, final-caption translation, October 5, 2026; Jicofo caption requests, September 2026, source code; Deepgram languages, checked October 6, 2026, official doc.

Open questions

No real multilingual meeting or provider credentials were tested for this guide. The complete cause and a reliable workaround for #635 remain unconfirmed. Per-speaker frontend language selection reaching bridge recognition needs testing; the proxy’s connection-level language option does not prove that propagation. Test Jigasi Whisper with differing spoken and caption languages before relying on that combination. [S14][S17][S18]

How do I set Jitsi caption languages and enable live translation?