Language Support
Omniflow’s language coverage depends on the channel and the underlying provider. This page is the canonical matrix.
Chat / text
| Family | Languages |
|---|---|
| DACH | German (DE-DE, DE-CH, DE-AT), Swiss German dialects (recognized via system prompt) |
| Romance | French (FR-FR, FR-CA), Italian, Spanish (ES-ES, ES-MX), Portuguese (PT-PT, PT-BR), Romanian, Catalan |
| English variants | en-US, en-GB, en-AU, en-IE, en-IN |
| Northern Europe | Dutch, Swedish, Norwegian, Danish, Finnish |
| Central / Eastern Europe | Polish, Czech, Slovak, Hungarian, Bulgarian, Russian, Ukrainian (Cyrillic supported in chat UI) |
| South / Southeast Asia | Hindi, Tamil, Telugu, Bengali, Bahasa Indonesia, Bahasa Malaysia, Thai, Vietnamese, Tagalog |
| CJK | Mandarin (Simplified + Traditional), Cantonese, Japanese, Korean |
| MENA | Arabic (modern standard + Gulf, Levantine, Maghreb dialects), Hebrew, Turkish, Farsi |
| Niche | Rhaeto-Romansch (Sursilvan, Vallader, Rumantsch Grischun) — limited but supported |
Chat UI is fully localized for EN, DE, FR, ES. Other languages are supported for content but the surrounding UI uses the closest of those four.
Voice — Realtime
Realtime voice agents inherit language coverage from the underlying provider:
| Provider | Strong | Adequate | Limited |
|---|---|---|---|
| OpenAI Realtime | EN, DE, FR, ES, IT, PT | NL, SV, JA, KO | Polish, Czech, Arabic |
| Gemini Live | EN, DE, FR, ES, IT, PT, JA, KO, ZH (Mandarin) | Bahasa, Hindi, Vietnamese, Thai | Cantonese, Cyrillic dialects |
| Cascaded (STT + LLM + TTS) | Anything STT/TTS supports — see below | — | — |
For high-fidelity multilingual voice, use Gemini Live as the primary and Cascaded as the fallback.
STT (cascaded)
Cascaded mode lets you bring your own STT and TTS, which expands language coverage. Supported STT providers:
| Provider | Coverage |
|---|---|
| OpenAI Whisper | 90+ languages incl. Arabic dialects, Cantonese, Bahasa. |
| Deepgram | 30+ languages with low latency; strong on EN, DE, FR, ES, JA, KO. |
| Google Speech-to-Text | 125+ languages; strong on CJK, Bahasa, Tamil. |
| Azure Speech | 100+ languages; strong on Arabic, Hebrew, Turkish. |
| Custom STT | OpenAI-compatible REST endpoint can be plugged in. |
TTS (cascaded)
| Provider | Coverage |
|---|---|
| ElevenLabs | 30+ languages with strong voice quality and cloning support. |
| Azure Neural TTS | 100+ languages, multiple regional voices. |
| AWS Polly | 60+ languages, fast, cheap. |
| Google Cloud TTS | 50+ languages with WaveNet quality. |
| OpenAI TTS | 6 voices, EN-strong, multilingual via translation. |
DTMF (touch-tone keypad)
DTMF input is supported in voice-to-voice flows for IVR-style interactions:
Agent: "For account balance press 1, for transfers press 2."
(DTMF tone detected)
Agent: "Looking up your balance..."Configure DTMF prompts in the agent’s prompt; Omniflow detects the tones from the audio stream and surfaces them as dtmf:<digit> events in the trace.
Speech Analytics language coverage
The Speech Analytics scoring grader works on every language the underlying STT supports. Rubric criteria descriptions can be authored per-language so the grader scores in the local language’s idiom.
| Language | Built-in rubric? | Custom rubric? |
|---|---|---|
| EN, DE, FR, ES, IT, PT, NL | âś… | âś… |
| Polish, Czech, Hungarian | âś… | âś… |
| JA, KO, ZH | Limited (idiomatic empathy detection less strong) | âś… |
| Arabic | Limited | âś… |
| Bahasa Indonesia / Malaysia | âś… (recently added) | âś… |
Mixed-language conversations
The agent auto-detects the customer’s language from the first turn and switches. For mixed-language emails (common in DACH and the Indian subcontinent), the Email Robot Assistant detects per-paragraph language and replies in the dominant one — or in a configured preferred language.
What “supported” means
| Tier | Definition |
|---|---|
| Strong | Built-in rubric exists; voice quality is production-ready; multiple QA-customer signoffs. |
| Adequate | Voice quality is good for support workloads; rubric needs minor tuning. |
| Limited | Available but with known caveats; recommend pilot before production. |
If your language isn’t listed and you have a use case, contact your account team — most languages can be enabled within 2-4 weeks of customer-led tuning.
Open in Omniflow
Related
| If you want to… | Go to |
|---|---|
| Pick a voice model | Voice Models |
| Set the workspace default | AI Infrastructure |
| Read about the Email Robot | Email Robot Assistant |