Skip to main content
QwenCloud home page
Docs
Speech-to-speech

Speech-to-speech overview

Choose a model for voice conversation, speech translation, or simultaneous interpretation.

Migrate from closed-source models

If you use an OpenAI or Gemini real-time speech model, use the following table to select a QwenCloud model for the same use case.
Closed-source examplesQwenCloud recommendation
Real-time conversationOpenAI GPT-Live 1, Gemini 3.8 Liveqwen-audio-3.1-realtime-plus
Cost-sensitive conversationOpenAI GPT-Realtime-2.1 Miniqwen-audio-3.1-realtime-flash
Real-time translationGemini 3.5 Live Translate (Preview)qwen3.8-livetranslate-flash-realtime
This page covers speech-to-speech. For visual understanding, audio/video analysis, and content moderation, see Omni-modal. For tasks that require reasoning and text output, such as video analysis and content labeling, see Qwen3.8-Omni-Flash.

S2S (speech-to-speech) vs. pipeline

There are two approaches to building voice applications:
S2SPipeline (ASR + LLM + TTS)
LatencyLow -- single-model stream processingHigher -- three-stage serial processing
Audio understandingEnd-to-end -- perceives tone and emotion and responds accordinglyConverts to text before processing, losing subtle audio cues
  • Use S2S for low latency, audio-aware responses, and interactive conversation.
  • Use a pipeline when you want to select ASR, LLM, and TTS models independently.
This page covers the S2S single-model approach (Omni, Qwen-Audio Realtime, and Livetranslate series). For the pipeline approach, select each component separately:

Real-time or file mode?

  • Real-time: Voice assistants, call centers, and simultaneous interpretation. Streams audio input and speech output.
Realtime API provides WebSocket, WebRTC, and AOQ access. For model support and protocol selection, see Realtime API overview.
  • File mode (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 use case (S2S single-model approach)

All use cases below use the S2S single-model approach. For the pipeline approach, use the ASR, LLM, and TTS guides linked above.
Use caseRecommended modelAPI
Realtime audio/video conversationqwen3.8-omni-flash-realtimeRealtime API
Voice assistants and customer-service conversationsqwen-audio-3.1-realtime-plusRealtime API
Cost-sensitive conversationsqwen-audio-3.1-realtime-flashRealtime API
Simultaneous interpretation and live translationqwen3.8-livetranslate-flash-realtimeRealtime API
Video dubbing and podcast translationqwen3-livetranslate-flashChat Completions
Semantic VAD voice assistants and smart customer service (with Function Calling support)qwen-audio-3.1-realtime-plusRealtime API

Companion capabilities of the S2S single-model approach

The following sections cover tool calling, web search, and text reasoning in voice applications.

Function calling

To query knowledge bases, check schedules, or trigger workflows based on audio/video content, use Qwen3.8-Omni-Flash-Realtime, or Qwen-Audio Realtime. For current information with spoken responses, use Qwen3.8-Omni-Flash-Realtime or Qwen3.5-Omni-Realtime for real-time conversations, or Qwen3.5-Omni (Chat Completions, both Plus and Flash) for file-based calls. The model decides autonomously whether to search. Qwen-Audio Realtime 3.1 Plus/Flash and 3.0 Plus/Flash also support web search through enable_search. All four Qwen-Audio Realtime models can use web search and Function Calling in the same session.
Not supported by Qwen3-Omni-Flash or the Livetranslate model.

Thinking mode

When answer quality matters more than latency, use qwen3.8-omni-flash (thinking enabled by default) or Qwen3.5-Omni and Qwen3-Omni-Flash over Chat Completions. The model reasons step by step before replying, which suits video analysis and batch labeling. Qwen-Audio Realtime does not support thinking mode.
Qwen3-Omni-Flash does not support voice generation in thinking mode.

Speech translation

The following model series support speech translation:
  • Qwen3.8-Livetranslate: Supports 60 source languages, speech output in 29 languages, and audio and image input. See Model information.
  • 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.
  • Qwen3.8-Omni-Flash-Realtime: Supports real-time speech translation and the same 36 speech output languages and dialects as Qwen3.5-Omni-Realtime. See the voice list for the languages supported by each voice.
  • Qwen3.5-Omni: 29 output languages and 7 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.
Use Livetranslate for ready-to-use translation. Choose Qwen3.5-Omni for offline speech output with web search and terminology injection.
LanguageQwen3.8 / 3.5-LivetranslateQwen3-LivetranslateQwen3.5-OmniQwen3-Omni-Flash
EnglishSupportedSupportedSupportedSupported
Chinese (Mandarin)SupportedSupportedSupportedSupported
CantoneseText-onlySupportedSupportedSupported
Sichuan dialectSupportedSupportedSupportedSupported
ShanghaineseSupportedSupported--Supported
Beijing dialectSupportedSupportedSupportedSupported
Tianjin dialectSupportedSupportedSupportedSupported
Nanjing dialect----SupportedSupported
Shaanxi dialect----SupportedSupported
Minnan dialect----SupportedSupported
FrenchSupportedSupportedSupportedSupported
GermanSupportedSupportedSupportedSupported
RussianSupportedSupportedSupportedSupported
ItalianSupportedSupportedSupportedSupported
SpanishSupportedSupportedSupportedSupported
PortugueseSupportedSupportedSupportedSupported
JapaneseSupportedSupportedSupportedSupported
KoreanSupportedSupportedSupportedSupported
ThaiSupportedText-onlySupportedSupported
IndonesianSupportedText-onlySupported--
VietnameseSupportedText-onlySupported--
ArabicSupportedText-onlySupported--
HindiSupportedText-onlySupported--
TurkishSupportedText-onlySupported--
FinnishSupported--Supported--
PolishSupported--Supported--
DutchSupported--Supported--
CzechSupported--Supported--
UrduSupported--Supported--
TagalogSupported--Supported--
SwedishSupported--Supported--
DanishSupported--Supported--
HebrewSupported--Supported--
IcelandicSupported--Supported--
MalaySupported--Supported--
NorwegianSupported--Supported--
PersianSupported--Supported--
GreekText-onlyText-only----
AfrikaansText-only------
AsturianText-only------
BelarusianText-only------
BulgarianText-only------
BengaliText-only------
BosnianText-only------
CatalanText-only------
CebuanoText-only------
EstonianText-only------
GalicianText-only------
GujaratiText-only------
CroatianText-only------
HungarianText-only------
JavaneseText-only------
KazakhText-only------
KannadaText-only------
KyrgyzText-only------
LatvianText-only------
MacedonianText-only------
MalayalamText-only------
MarathiText-only------
PunjabiText-only------
RomanianText-only------
SlovakText-only------
SlovenianText-only------
SwahiliText-only------
TajikText-only------
AzerbaijaniText-only------
UkrainianText-only------
"Supported" = speech + text output. "Text-only" = text output only, no speech.Qwen3.8-Omni-Flash-Realtime and Qwen3.5-Omni support 113 input languages and dialects.Qwen3.8 / 3.5-Livetranslate supports 60 languages (29 with audio and text, 31 text only).
The table lists the entry-point model in each series. To pin a dated version for regression testing or stability, see All models below.
ModelAPIUse cases
qwen3.8-omni-flashChat Completions / ResponsesAudio/video understanding and text generation
qwen3.8-omni-flash-realtimeRealtime APIRealtime audio/video conversation
qwen-audio-3.1-realtime-plus / qwen-audio-3.1-realtime-flashRealtime APIRealtime voice conversation
qwen3.5-omni-plus / qwen3.5-omni-flashChat CompletionsOffline speech output
qwen3.8-livetranslate-flash-realtimeRealtime APIRealtime translation
qwen3-livetranslate-flashChat CompletionsAudio/video file translation

All models

Qwen-Audio

ModelAPIInputFunction callingWeb searchThinking modeTranslation
qwen-audio-3.1-realtime-plusRealtime APIaudio, textSupportedSupported----
qwen-audio-3.1-realtime-flashRealtime APIaudio, textSupportedSupportedNot supported--
qwen-audio-3.0-realtime-plusRealtime APIaudio, textSupportedSupported----
qwen-audio-3.0-realtime-flashRealtime APIaudio, textSupportedSupported----

Qwen3.8-Omni

ModelAPIInputFunction callingWeb searchThinking mode
qwen3.8-omni-flashChat Completions / ResponsesText, audio, image, videoSupportedSupportedSupported
qwen3.8-omni-flash-realtimeRealtime APIText, audio, image, videoSupportedSupported--
qwen3.8-omni-flash returns text only, with thinking enabled by default, and suits audio/video understanding and content analysis; qwen3.8-omni-flash-realtime returns text and audio and also supports multichannel audio and MCP. For calling details, see audio/video file understanding and the real-time guide.

Qwen3.5-Omni

ModelAPIInputFunction callingWeb searchThinking mode
qwen3.5-omni-plus-realtimeRealtime APIText, audio, image, videoSupportedSupported--
qwen3.5-omni-plus-realtime-2026-03-15Realtime APIText, audio, image, videoSupportedSupported--
qwen3.5-omni-plusChat CompletionsText, audio, image, videoSupported (China North 2 (Beijing), text output)Supported--
qwen3.5-omni-plus-2026-03-15Chat CompletionsText, audio, image, videoSupported (China North 2 (Beijing), text output)Supported--
qwen3.5-omni-flash-realtimeRealtime APIText, audio, image, videoSupportedSupported--
qwen3.5-omni-flash-realtime-2026-03-15Realtime APIText, audio, image, videoSupportedSupported--
qwen3.5-omni-flashChat CompletionsText, audio, image, videoSupported (China North 2 (Beijing), text output)Supported--
qwen3.5-omni-flash-2026-03-15Chat CompletionsText, audio, image, videoSupported (China North 2 (Beijing), text output)Supported--

Qwen3-Omni

ModelAPIInputFunction callingWeb searchThinking mode
qwen3-omni-flash-realtime-2025-09-15WebSocketText, audio, image, video------
qwen3-omni-flashChat CompletionsText, audio, image, videoSupported--Supported

Qwen3.8-Livetranslate

ModelAPIInputLanguages
qwen3.8-livetranslate-flash-realtimeRealtime APIAudio, image60

Qwen3.5-Livetranslate

ModelAPIInputLanguages
qwen3.5-livetranslate-flash-realtimeRealtime APIAudio, images60
qwen3.5-livetranslate-flash-realtime-2026-05-19Realtime APIAudio60

Qwen3-Livetranslate

ModelAPIInputLanguages
qwen3-livetranslate-flashChat CompletionsAudio, video18
qwen3-livetranslate-flash-2025-12-01Chat CompletionsAudio, video18

What's next

API documentation by model series: