Skip to content

Languages

What Vela speaks, layer by layer. The UI, the spoken turn-by-turn, the neural voice, and on-device dictation are separate systems, so support differs per language - this table is the canonical list. Update it when a language lands or gains a layer.

Language Code App UI Spoken directions Vela voice (neural TTS) Dictation (voice search)
English en ✅ ✅ ✅ (US + British voices) ✅
French fr ✅ ✅ ✅ ✅
German de ✅ ✅ ✅ ✅
Spanish es ✅ ✅ ✅ (Spain + Mexico voices) ✅
Italian it ✅ ✅ ✅ ✅
Portuguese (Portugal) pt ✅ ✅ ✅ (Brazilian voice) ✅
Portuguese (Brazil) pt-BR (resources in values-pt-rBR) ✅ (contributed by Netocon, 2026-09-28) ✅ (shares the Portuguese table) ✅ ✅
Dutch nl ✅ ✅ ✅ ✅
Russian ru ✅ ✅ ✅ ✅
Polish pl ✅ ✅ ✅ ✅
Swedish sv ✅ ✅ ✅ ✅
Ukrainian uk ✅ ✅ ✅ ✅
Hungarian hu ✅ (contributed by Zsolt Laszlo Kaiser, 2026-09-13) ✅ ✅ (Anna) ✅
Chinese (Simplified) zh ✅ ✅ ✅ (Mandarin voice) ✅
Chinese (Traditional) zh-TW ✅ ✅ ✅ (shares the Mandarin voice) ✅
Japanese ja ✅ ✅ ❌ system TTS* ✅
Hebrew he (resources in values-iw) ✅ (first RTL locale) ✅ ❌ system TTS* ✅

* Piper/espeak-ng has no Japanese or Hebrew phonemizer, so there's no downloadable Vela voice for those two. Spoken directions route to the phone's own system TTS in that language instead; if the system has no such voice either, nav stays silent rather than mangling it, and a hint points at the voice settings.

The App UI column means the language has its own string file, not that every string in it is translated. New features land in English first and show in English until someone fills them in; all sixteen files were brought fully up to date with English on 2026-09-30 (the 54 strings added since 2026-09-28: region parts, saved routes and pins, per-place icons, the map theme, end-navigation confirm), translated per language to match each file's own register and terms. python3 tools/check-translations.py lists the missing keys per language (and fails only on placeholder drift); values-en-rGB is left out of that list, since it only ever carries the few words that differ.

Some context on the columns:

  • App UI - every user-facing string, from Settings to the nav banner to the foreground-service notification. Google place content (categories, hours, open/closed) also arrives localized, since the scrape's hl= follows the app language. Place NAMES, street names and reviews are data and are never translated.
  • Spoken directions - generated from per-language grammar templates (NavStrings in :core), not word-swapped, so cases and word order are right. Distances follow the imperial/metric setting.
  • Vela voice - the downloadable on-device neural voice (Piper). The voice library pairs the app language to a matching voice and nudges a download if you switch to a language whose voice isn't installed.
  • Dictation - the search-bar mic transcribes on-device. The default is a multilingual Whisper model, pinned to the app language for every language in the table. Settings → Search also offers two alternative engines you can download and switch to: SenseVoice (English, Chinese, Japanese, Korean, Cantonese) and Moonshine (English only) - faster/more accurate for those languages, but Whisper stays the default so nothing regresses.

The language setting is in Settings → Appearance: "Follow system language" is on by default, and turning it off shows a picker with English and the fifteen translations, each under its own name (Deutsch, 日本語, עברית and so on).

Adding a language

The moving parts, in the order they matter:

  1. app/src/main/res/values-<code>/strings.xml - the full UI string set (translated from values/strings.xml, the English source). Placeholder types must match English exactly; CI validates this. Count-bearing strings are <plurals> - give the language the plural forms it actually needs.
  2. AppLocale.SUPPORTED + its endonym map (app/ui/AppLocale.kt) - registers the language in the picker and everywhere the app language flows (dictation pinning, the hl= scrape parameter, voice pairing).
  3. A NavStrings table in core/i18n/ - the spoken-direction grammar templates.
  4. The open/closed status keyword table (SearchParser) and the transit-category words (calibration) - the two spots that match localized Google TEXT to make a decision.
  5. The other per-language word tables, each a plain map in :core: the voice-command vocabulary (search/QueryIntent) and its example phrases for Settings > Search (search/VoiceCommandExamples), the review page's labels (data/ReviewWords), and the generic business words the same-business rule ignores (util/PlaceNames, whose union is mirrored in tools/place-generic-words.txt for the places bake).
  6. Optional: a Piper voice for the language in PiperCatalog if one exists upstream.

QueryIntentTest compares the voice-command tables with the app's language list and fails when one is missing. PlaceStatusTest (the status-table languages) and SpokenRoadNamesTest (every nav table) pin their own lists, so add the new code there too, then run ./gradlew :core:test.

One literal is deliberately still English-only because it doubles as a logic key: the "Open"/"Closed" word on a status line Vela computes itself from the hours (used only when Google sent no status text), which the status coloring parses. It localizes once display text is split from that key. The category chips and the review sort and tab labels used to be in this list; their labels are translated now and only the query or click key behind them stays English.

Weblate

Not live yet, and there is no Weblate project to sign in to. Hosted Weblate is free for open source once a project is three months old; Vela passed that mark on 2026-09-15, so the only thing missing is the application itself, which has not been made. Until a project is accepted, translations come in as ordinary pull requests: see TRANSLATING.md. New strings go into the English base file only, and a missing translation falls back to English. The values-<code>/strings.xml layout, the plurals and the placeholder check in CI are what Weblate needs, and all three are already in place.