Speech-to-speech models

​

Migrate from closed-source models

Map your current OpenAI Realtime or Gemini Live setup to an equivalent QwenCloud model. This page covers speech-to-speech. For visual understanding, audio/video analysis, or content moderation, see the Omni-modal documentation.

​

S2S vs pipeline

Two ways to build voice apps:

• Use S2S when interactive conversation, low latency, and audio-aware responses matter. Continue reading this page.
• Use Pipeline when you need voice customization or want to select ASR, LLM, and TTS models independently.

This page covers the S2S single-model approach (Omni and Livetranslate series). For the pipeline approach, select each component separately:

• ASR (speech recognition): Speech-to-text
• LLM (language model): Text generation
• TTS (speech synthesis): Text-to-speech

​

Real-time or file-based?

• Real-time (WebSocket) — Voice assistants, call centers, and simultaneous interpretation. Streams audio input and speech output.
• File-based (HTTP) — Higher latency but better quality. Ideal for video dubbing, podcast translation, and offline processing. Also supports function calling, web search, thinking mode, and video context (see Companion capabilities below).

​

Choose a model by scenario (S2S single-model route)

All use cases below use the S2S single-model approach. For the pipeline approach, use the ASR, LLM, and TTS guides linked above.

​

Companion capabilities of S2S models

Qwen3.5-Omni and Qwen3-Omni provide these capabilities natively. Qwen-Audio Realtime supports function calling only; it does not support web search or thinking mode. With a pipeline, equivalent functionality comes from individual components (typically the LLM).

​

Function calling

The model can query knowledge bases, check schedules, or trigger workflows based on what it hears and sees. Use Qwen3.5-Omni (WebSocket or HTTP), Qwen3-Omni (HTTP only), or Qwen-Audio Realtime (WebSocket).

​

Web search

Retrieves real-time information for current events, stock prices, weather, and similar queries. Available in Qwen3.5-Omni (WebSocket or HTTP, both Plus and Flash). The model decides autonomously whether to search. Qwen-Audio Realtime does not support this capability.

​

Thinking mode

Use Qwen3-Omni (HTTP) when answer quality outweighs latency. Reasons step by step before replying, ideal for video analysis and batch labeling. Qwen-Audio Realtime does not support this capability.

​

Translation

All model families can translate speech:

• Qwen3.5-Livetranslate — 60 languages (29 with audio+text output, 31 text-only). Covers Chinese, English, French, German, Russian, Japanese, Korean, Spanish, Portuguese, Arabic, and more.
• Qwen3-Livetranslate — 18 languages and 5 Chinese dialects (~3 s latency). File mode accepts video input for context-aware translations. 7 languages produce text-only output.
• Qwen3.5-Omni — 29 output languages and 8 Chinese dialects. Strong audio/video understanding and web search. Inject terminology and domain context via system prompt. Real-time and file modes.
• Qwen3-Omni-Flash — 11 output languages and 8 Chinese dialects. Inject terminology and domain context via system prompt. Real-time and file modes, at lower cost.

Supported languages

Language	Qwen3.5-Livetranslate	Qwen3-Livetranslate	Qwen3.5-Omni	Qwen3-Omni-Flash
English	✓	✓	✓	✓
Chinese (Mandarin)	✓	✓	✓	✓
+ Cantonese	Text only	✓	✓	✓
+ Sichuanese	✓	✓	✓	✓
+ Shanghainese	✓	✓	✓	✓
+ Beijing	✓	✓	✓	✓
+ Tianjin	✓	✓	✓	✓
+ Nanjing	—	—	✓	✓
+ Shaanxi	—	—	✓	✓
+ Hokkien	—	—	✓	✓
French	✓	✓	✓	✓
German	✓	✓	✓	✓
Russian	✓	✓	✓	✓
Italian	✓	✓	✓	✓
Spanish	✓	✓	✓	✓
Portuguese	✓	✓	✓	✓
Japanese	✓	✓	✓	✓
Korean	✓	✓	✓	✓
Arabic	✓	Text only	✓	—
Thai	✓	Text only	✓	✓
Vietnamese	✓	Text only	✓	—
Indonesian	✓	Text only	✓	—
Turkish	✓	Text only	✓	—
Hindi	✓	Text only	✓	—
Malay	✓	—	✓	—
Dutch	✓	—	✓	—
Urdu	✓	—	✓	—
Norwegian	✓	—	✓	—
Swedish	✓	—	✓	—
Danish	✓	—	✓	—
Hebrew	✓	—	✓	—
Finnish	✓	—	✓	—
Polish	✓	—	✓	—
Icelandic	✓	—	✓	—
Czech	✓	—	✓	—
Tagalog	✓	—	✓	—
Persian	✓	—	✓	—
Greek	Text only	Text only	—	—
Afrikaans	Text only	—	—	—
Asturian	Text only	—	—	—
Belarusian	Text only	—	—	—
Bulgarian	Text only	—	—	—
Bengali	Text only	—	—	—
Bosnian	Text only	—	—	—
Catalan	Text only	—	—	—
Cebuano	Text only	—	—	—
Estonian	Text only	—	—	—
Galician	Text only	—	—	—
Gujarati	Text only	—	—	—
Croatian	Text only	—	—	—
Hungarian	Text only	—	—	—
Javanese	Text only	—	—	—
Kazakh	Text only	—	—	—
Kannada	Text only	—	—	—
Kyrgyz	Text only	—	—	—
Latvian	Text only	—	—	—
Macedonian	Text only	—	—	—
Malayalam	Text only	—	—	—
Marathi	Text only	—	—	—
Punjabi	Text only	—	—	—
Romanian	Text only	—	—	—
Slovak	Text only	—	—	—
Slovenian	Text only	—	—	—
Swahili	Text only	—	—	—
Tajik	Text only	—	—	—
Azerbaijani	Text only	—	—	—
Ukrainian	Text only	—	—	—

"✓" = speech + text output. "Text only" = text output only, no speech.Qwen3.5-Omni supports 113 input languages/dialects total. See full list for details.Legacy qwen-omni-turbo supports Chinese and English only.

​

Recommended models

The table lists the entry-point model in each series. To pin a dated version for regression testing or stability, see All models below.

​

All models

​

Next steps

API documentation by model series:

• Qwen3.5-Omni / Qwen3-Omni (WebSocket, real-time) → Real-time multimodal speech
• Qwen3.5-Omni / Qwen3-Omni (HTTP, file-based) → Multimodal speech
• Qwen3.5-Livetranslate (WebSocket, real-time) → Real-time translation
• Qwen3-Livetranslate (HTTP, file-based) → File-based translation

​

Learn more