🌍 Language Support

25 UI locales · 21 TTS languages · 21 AAC offline dictionaries. Synalux is built for multilingual practices and the families they serve. Below is the honest coverage matrix per surface.

Coverage Matrix

Language	UI translation	Premium TTS (Deepgram)	Fallback TTS (Azure)	Offline TTS (Kokoro)	AAC offline dict (500 words)	Cloud autocorrect (Gemini)
🇺🇸 English (US/GB/AU)	✅	✅ 6 voices	—	✅	✅	✅
🇪🇸 Spanish (ES/MX)	✅	✅ Carmen / Diego	—	✅	✅	✅
🇫🇷 French	✅	✅ Camille / Lucas	—	✅	✅	✅
🇩🇪 German	✅	✅ Hans / Lena	—	—	✅	✅
🇧🇷🇵🇹 Portuguese	✅	✅ Luana	—	✅	✅	✅
🇮🇹 Italian	✅	✅ Giulia	—	—	✅	✅
🇳🇱 Dutch	✅	✅ Lotte	—	—	✅	✅
🇵🇱 Polish	✅	✅ Zofia	—	—	✅	✅
🇯🇵 Japanese	✅	✅ さくら	—	✅	✅	✅
🇨🇳 Chinese (Mandarin)	✅	✅ 美	—	✅	✅	✅
🇨🇳 Chinese (zh-TW)	—	—	✅ 曉臻	—	—	✅
🇨🇳 Chinese (zh-HK)	—	—	✅ 曉曼	—	—	✅
🇰🇷 Korean	✅	✅ 지수	—	—	✅	✅
🇷🇺 Russian	✅	✅ Аня	—	—	✅	✅
🇷🇴 Romanian	✅	— (subsidized via Azure on free tier)	✅ Alina	—	✅	✅
🇺🇦 Ukrainian	✅	— (subsidized via Azure on free tier)	✅ Поліна	—	✅	✅
🇸🇦 Arabic (RTL)	✅	✅ ليلى	—	—	✅	✅
🇮🇱 Hebrew (RTL)	✅	✅ נועה	—	—	✅	✅
🇮🇳 Hindi	✅	✅ आन्या	—	—	✅	✅
🇻🇳 Vietnamese	✅	—	✅ Hoài Mỹ + Nam Minh	—	✅	✅
🇵🇭 Filipino (Tagalog)	✅	—	✅ Blessica + Angelo	—	✅	✅
🇹🇷 Turkish	✅	—	✅ Emel + Ahmet	—	✅	✅
🇮🇩 Indonesian	✅	—	✅ Gadis + Ardi	—	✅	✅

Global Dashboard — Multi-Language Support

Counts at a Glance

Surface	Languages	Notes
UI translations	25 locales (100% coverage each)	en×3, es×2, pt×2 + 18 single-region locales
Premium TTS (Deepgram)	14 base languages, ~24 voices	en, es, fr, de, pt, it, nl, pl, ja, zh, ko, ru, he, ar, hi
Fallback TTS (Azure)	8 languages, ~14 voices	ro, uk, zh-tw, zh-hk, vi, tl, tr, id
Offline neural TTS (Kokoro)	6 languages	en, es, fr, pt, ja, zh — bundled WASM
AAC offline dictionary	21 languages × 500 words each	Index-aligned across langs
Whisper transcription	~99 languages	Whisper's full coverage; quality varies by language
Cloud autocorrect (Gemini Flash-Lite)	~100+ languages	Bench-validated for en/ro/ru/es/uk/pl

How the Layered Fallback Works

If your language doesn't have premium TTS, you still get a voice. If you don't have Whisper-trained data, you still get cloud transcription. If you're offline, you still get the AAC dictionary.


TTS:        Deepgram (Tier 1) → Kokoro WASM (Tier 1.5) → Web Speech API (Tier 2) → espeak-ng WASM (Tier 3)
Translation: Offline dictionary → Cloud Gemini Flash → original text passthrough
Autocorrect: Local prism-coder (when running) → Cloud Gemini Flash-Lite → original text passthrough



See Voice TTS Architecture, Translation, Transcription for the full per-surface design.



Subsidized Languages (Free Tier)

For users in regions where Synalux serves underserved populations, Synalux absorbs Deepgram Aura cost on the free tier so users get premium voices at no charge:

ro Romanian · uk Ukrainian · ru Russian · de German · ko Korean · ar Arabic



Without this subsidy, users in these regions would either pay or use Tier 1.5/2/3 fallbacks.



Adding a Language

The translation pipeline is automated. To add a 26th UI locale:

Add the locale code to Locale type + LOCALE_LABELS in portal/src/lib/i18n.ts.

Add an empty '': {} entry in the translations map.

Run the bulk-translation script (one-shot, not committed) — pulls all 257 keys from en-US, batch-translates via Gemini Flash-Lite (~$0.005), writes back the locale block.

Add a voice entry to portal/src/shared/voice-catalog.ts` (Deepgram if available; else Azure fallback).

Run the AAC dict script for the matching prism-aac repo (translates 500 entries via Gemini, ~$0.005).

tsc + commit.

End-to-end: ~10 minutes per new language, ~$0.01 in API spend.

Roadmap

Languages frequently requested but not yet supported:

Greek (el), Czech (cs), Hungarian (hu), Swedish (sv), Norwegian (no), Danish (da), Finnish (fi)

Bengali (bn), Tamil (ta), Telugu (te), Marathi (mr), Punjabi (pa) — South Asian sub-languages

Swahili (sw), Yoruba (yo), Hausa (ha) — African languages

If you need a specific language for your practice, contact us — turnaround is typically 1-2 weeks for adding a new locale across all surfaces.