Voice Frontends
The voice frontend handles realtime conversation; the backend Agent executes work. You can choose them independently.
Put these settings in the config.env shown by qwenaudio config, or select a service and enter credentials in Desktop's voice frontend settings.
Choose a Service
| Service | Provider value | Required configuration or preparation | Guide |
|---|---|---|---|
| Qwen Audio / Omni 3.5 / Omni 3.8 | dashscope (default) | DASHSCOPE_API_KEY; Omni 3.8 also requires a workspace-specific QWEN_AUDIO_REALTIME_BASE_URL | Audio / Omni vision |
| StepAudio 3 | stepfun | STEPFUN_API_KEY | StepFun |
| OpenAI Realtime | gpt-live | OPENAI_API_KEY | GPT-Live |
| Gemini Live | google-live | GOOGLE_API_KEY | Google Live |
| Doubao Seeduplex | doubao-seeduplex | DOUBAO_API_KEY | Configure its model, voice, and endpoint below |
| Hugging Face speech-to-speech | speech-to-speech | Start the service; default: ws://127.0.0.1:8765/v1/realtime | Local model pipeline |
| MiniCPM-o 4.5 | minicpm-o | Start the service; default: ws://127.0.0.1:8006/v1/realtime?mode=audio | Audio/video modes and limits |
For the default frontend:
QWEN_AUDIO_REALTIME_PROVIDER=dashscope
DASHSCOPE_API_KEY=your-key
QWEN_AUDIO_REALTIME_MODEL=qwen-audio-3.0-realtime-plusTo switch to StepFun, select it and supply its own credentials:
QWEN_AUDIO_REALTIME_PROVIDER=stepfun
STEPFUN_API_KEY=your-stepfun-keyYou can keep all provider settings in the same file. Switch using QWEN_AUDIO_REALTIME_PROVIDER; the Gateway does not reuse another provider's key, model, or voice.
Model, Voice, and Endpoint
| Provider | Model | Voice | Endpoint |
|---|---|---|---|
| DashScope | QWEN_AUDIO_REALTIME_MODEL | Audio: QWEN_AUDIO_REALTIME_VOICE; Omni: QWEN_OMNI_REALTIME_VOICE | QWEN_AUDIO_REALTIME_BASE_URL |
| StepFun | STEPFUN_REALTIME_MODEL | STEPFUN_REALTIME_VOICE | STEPFUN_REALTIME_URL |
| GPT-Live | GPT_LIVE_REALTIME_MODEL | GPT_LIVE_REALTIME_VOICE | GPT_LIVE_REALTIME_URL |
| Google Live | GOOGLE_LIVE_REALTIME_MODEL | GOOGLE_LIVE_REALTIME_VOICE | GOOGLE_LIVE_REALTIME_URL |
| Doubao Seeduplex | DOUBAO_SEEDUPLEX_REALTIME_MODEL | DOUBAO_SEEDUPLEX_REALTIME_VOICE | DOUBAO_SEEDUPLEX_REALTIME_URL |
| speech-to-speech | Configure upstream | Configure upstream | SPEECH_TO_SPEECH_REALTIME_URL |
| MiniCPM-o | Configure upstream | Configure upstream | MINICPM_O_REALTIME_URL |
Empty endpoint and model fields use provider defaults, except Qwen3.8 Omni, which requires a workspace-specific endpoint. For self-hosted services behind Bearer authentication, set SPEECH_TO_SPEECH_AUTH_TOKEN or MINICPM_O_AUTH_TOKEN. Aliases and protocol details are listed in each service guide.
Current built-in DashScope model profiles:
| Model ID | Realtime input |
|---|---|
qwen-audio-3.0-realtime-plus (default) | Text, audio |
qwen-audio-3.0-realtime-flash | Text, audio |
qwen3.5-omni-flash-realtime | Text, audio, live visual frames |
qwen3.5-omni-plus-realtime | Text, audio, live visual frames |
qwen3.8-omni-flash-realtime | Text, audio, live visual frames |
All profiles support tool calls. Live vision also requires client and transport support; see Visual Input.
Apply and Verify
- Click Apply in Desktop. For a terminal Gateway, stop and restart it. For an installed background service, run
qwenaudio gateway restart. - Connect a client and check that the voice frontend is connected.
- Speak and confirm that audio plays. If you need tools, also test search or a simple backend request.
To inspect the configured provider and supported models:
qwenaudio config show
qwenaudio config set --realtime-model qwen-audio-3.0-realtime-flashconfig set changes the current provider's model; it neither switches providers nor restarts the Gateway. Remote clients use the remote Gateway's settings, not local model settings.
Service Differences
- MiniCPM-o: the current adapter does not support text input, structured tool calls, proactive replies, or conversation context restoration. It supports voice/visual chat, not backend orchestration.
- speech-to-speech: recognition languages, voices, and tool-call quality depend on the STT, LLM, and TTS you configure.
- Google Live: the current adapter does not restore conversation context into a reconnected upstream session. Visible chat history does not mean the model received that history.
- Other supported features and limits are documented in the service-specific guides.
Configure frontend tools separately: Search, MCP, OpenAPI.
Older Configuration
Use provider-specific fields for new configurations. QWEN_AUDIO_REALTIME_API_KEY and QWEN_AUDIO_REALTIME_ENDPOINT are no longer process-environment overrides. Older files are converted when read; if conversion reports a conflict, keep the intended provider field and remove the obsolete field. See configuration priority.