Voice & Video: Input, Conversation, and Creation
Find dictation, live voice, read-aloud, meeting transcription, and image and video creation.
Configure: Settings → Media Services. Use voice in chat; meetings and Canvas have their own workspaces.
What is in this category?
| Feature | Purpose | Where to start |
|---|---|---|
| Voice input / Dictation | Turn speech into chat text | Enable voice input, then use the composer microphone or dictation shortcut |
| Live voice conversation | Speak with the assistant and interrupt naturally | Media Services → Live voice conversation → Open voice orb |
| Read-aloud | Listen to AI replies and choose a voice | Media Services → Voice output |
| Voice wake | Start dictation with a wake phrase | Media Services → Voice wake, desktop only |
| Meeting recording and transcription | Record or import audio, create text and minutes | Meetings |
| Image generation | Create images from text and references | Canvas |
| Video generation | Choose models, references, aspect ratio, and duration | Canvas video guide |
These features use separate services, so configure each one you need.
Start a voice conversation
- Check the voice service and microphone in Settings → Media Services.
- For dictation, enable voice input, then use the composer microphone or configured shortcut.
- For live conversation, click Open voice orb and follow the prompt to start speaking.
- Test one short sentence to check audio input, recognition, and replies.
- Stop recording or end the voice session when finished.
Result: dictation produces recognized text; live conversation produces a spoken reply. Some dictation modes send after releasing the shortcut or sustained silence. Check the current mode's sending behavior first.
Read-aloud and wake phrases
Choose a voice in Media Services and enable Automatically read AI replies aloud when desired.
For voice wake, first verify that the desktop client's local model is available, then set a wake phrase. Wake detection runs locally. Subsequent dictation or conversation processes audio using its selected voice service.
Create an image or video
- Start a Canvas project and add a prompt and reference material.
- Choose an image or video generation node and check the model's supported inputs.
- Set the available aspect ratio, count, resolution, or duration options.
- Review the cost and click Generate yourself.
- Wait for completion, preview the result, then save or export it.
The assistant can organize references, improve prompts, and prepare nodes. You manually start each Canvas model generation. See the Canvas guide.
Common problems
| Symptom | Check first |
|---|---|
| No microphone input | Selected microphone, system recording permission, and other apps using it |
| Dictation works but live voice does not | Live voice service, account state, and feature access |
| Replies are not read aloud | Read-aloud switch, selected voice, speakers, and service status |
| Meeting recording captures only one side | Microphone and system audio selection; make a short test recording |
| Video model unavailable or generation fails | Model capability, input formats, parameters, quota, and task status |
See Privacy & Local Data for audio and asset handling, and Account & Usage for usage information.