Speech-to-speech models

​

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 custom voices or want to mix-and-match the best ASR, LLM, and TTS for each stage.

This page covers the S2S single-model route (Omni, Livetranslate). For the Pipeline route, choose your three components 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) — Use for live voice interfaces: voice assistants, call centers, simultaneous interpretation. Audio streams in, speech streams out. Model names contain -realtime.
• File-based (HTTP) — Use when you can trade latency for better results: video dubbing, podcast translation, offline content processing. File-based mode also unlocks function calling, web search, thinking mode, and video context (see "Companion capabilities" below).

​

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

The following scenarios are for the S2S single-model route. For the Pipeline route, choose components from the ASR / LLM / TTS docs linked above.

​

Companion capabilities of S2S models

The following capabilities are provided directly by the Qwen3.5-Omni / Qwen3-Omni models in the S2S single-model route. In the Pipeline route, these capabilities need to be supported by the respective LLM or other components.

​

Function calling

Let the model take actions based on what it hears and sees — check a knowledge base, query a schedule, trigger a workflow. Supported by Qwen3.5 Omni (WebSocket and HTTP) and Qwen3 Omni (HTTP).

​

Web search

Let the model retrieve real-time information to answer questions about current events, stock prices, weather, and more. Supported by Qwen3.5 Omni (HTTP and WebSocket), including Plus and Flash variants. The model autonomously decides whether to search.

​

Thinking mode

Use qwen3-omni-flash (HTTP) when answer quality matters more than latency. The model reasons step-by-step before responding — useful for video analysis, batch tagging, and complex Q&A.

​

Translation

All model families can translate speech:

• Qwen3.5-Livetranslate — 60 languages (29 audio+text, 31 text-only), ~3-second latency, out of the box. WebSocket realtime only.
• Qwen3-Livetranslate — 18 languages + 5 Chinese dialects, ~3-second latency, out of the box. File-based variant accepts video for context-aware accuracy. 7 languages output text only (no audio).
• Qwen3.5-Omni — 29 output languages + 7 Chinese dialects. Superior audio-video understanding and web search. Inject terminology and domain context via system prompt. Both realtime and file-based.
• Qwen3-Omni-Flash — 11 output languages + 8 Chinese dialects. Inject terminology and domain context via system prompt for specialized fields. Both realtime and file-based. 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	—	—	—

✓ = audio + text output. "Text only" = no audio output for that language. Qwen3.5-Livetranslate supports 60 languages total (29 audio+text, 31 text-only).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 below lists the common entry point for each family. For pinned date versions (for version regression or stability), see "All models" below.

​

All models

​

Next steps

After choosing a model, refer to the corresponding usage guide:

• 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