Multilingual conversation
This feature provides single-language conversations for popular languages across Europe, the Americas, and Asia. Select a language when you create a multimodal application, and the system configures the most suitable speech models, prompts, and features automatically.
Supported languages
Language | Application type | Speech recognition | Speech synthesis | LLM conversation - detailed features |
Chinese (Mandarin) | Multimodal application Voice application | Supported | Supported |
|
English | Supported | Supported |
| |
French, German, Spanish, Italian, Russian, Portuguese, Korean, Japanese, Thai, Indonesian, Malay | Multimodal application | Supported (hotwords not supported) | Supported |
|
Cantonese | Supported (hotwords not supported) | Supported | ||
Vietnamese, Filipino | Supported (hotwords not supported) | Currently requires integration with a third-party model | ||
Arabic, Hindi, Turkish, Ukrainian, Czech, Danish, Finnish, Icelandic, Norwegian, Polish, Dutch, Swedish | Supported (hotwords not supported) | Currently requires integration with a third-party model |
Usage notes
- Only single-language conversations are supported; mixed-language conversations are not supported (except Chinese-English code-switching).
- To offer multiple languages on the same hardware device, create a separate application for each language and switch the application ID through system settings to deliver the conversation capability of the corresponding language to end users.
- Billing for the above languages follows the same logic as Chinese.
ASR and TTS model selection recommendations
-
The speech models available and recommended for each language are listed in the following table:
- An empty cell means that no speech model is currently available for that language, and you must integrate a third-party model.
Language | Recommended ASR model (performance first) | Available ASR models | Recommended TTS model & timbre (performance first) | Available TTS models |
Chinese |
|
|
|
|
English |
|
|
|
|
Japanese |
|
|
|
|
Korean |
|
|
|
|
French |
|
|
|
|
German |
|
|
|
|
Italian |
|
|
|
|
Spanish |
|
|
|
|
Portuguese |
|
|
|
|
Russian |
|
|
|
|
Thai |
|
|
|
|
Indonesian |
|
|
|
|
Filipino |
|
| ||
Cantonese |
|
|
|
|
Arabic |
|
| ||
Hindi |
|
| ||
Turkish |
|
| ||
Ukrainian |
|
| ||
Czech |
|
| ||
Danish |
|
| ||
Finnish |
|
| ||
Icelandic |
|
| ||
Norwegian |
|
| ||
Polish |
|
| ||
Dutch | Currently requires integration with a third-party model | Currently requires integration with a third-party model | ||
Swedish |
|
|
Configuration
-
In the console, click to create a "Multimodal Application".
-
Select the desired language and category to create the application. The system will provide the most suitable models, prompts, and feature combinations.
In this example, for Conversation language select English, and for the scenario template select AI Glasses.
-
Adjust the prompt and other content based on your business scenario. Features unavailable for the current language are hidden.
On the Speech Interaction tab, you can configure speech recognition (such as Fun-ASR Real-time Speech Recognition) and speech synthesis (such as the CosyVoice-v3-Flash Large Model), as well as options such as hotwords, instant correction, and conversation interruption (barge-in at any time/full-duplex, or click-to-interrupt/half-duplex). On the Understanding and Generation tab, you can configure options such as intent recognition, the text model (such as Qwen3.6-Plus), follow-ups, the knowledge base, web search, and long-term memory. The page also contains the Skills and Agent tabs for further configuration.
-
Click "Run Now", browse the available timbre list in the timbre selection area of the run panel on the right, and select a suitable timbre.
-
Test the conversation on the right, and proceed with subsequent steps such as publishing, development integration, and purchase.