Voice input and text selection

Get one speech channel working first

1. Open Voice input and add a supported account source, or configure a separate speech-recognition API model. An ordinary chat model does not necessarily support speech recognition.
2. Check the speech API URL, model and authorization. You can configure up to three different channels: the first is primary and the others are backups. Drag them to change their order.
3. Select the default microphone or a specific device, run the test, and check that the level changes as you speak. Stop the test when you finish.
4. Try recognizing a short sentence first. Once the text looks right, set a shortcut that does not conflict with your usual software.

Use it in the input field where you are working

1. First place the cursor in a chat box, editor or another text input field.
2. Start recording with the microphone control or a shortcut. Shortcuts support both hold-to-talk and toggle modes.
3. After you stop recording, wait for recognition and insertion to finish. Check the inserted text before deciding whether to send it.
4. If focus changes or the system prevents insertion, copy the retained transcription and paste it manually. You can also replay the recording when needed.

What the Voice Island shows

The Voice Island shows recording, recognition, insertion and completion states. Adjust its theme, colors and size to suit your desktop. It serves a different purpose from the Usage Island: the Voice Island follows the input process, while the Usage Island shows account and model usage.

Translate or ask about selected text

Configure an available text model first, then enable selection translation or questions. After selecting text, start an action near the selection and choose the target translation language. Each selection is limited to 16,000 characters; split longer text into sections.

If the system cannot retrieve the selection, copy it first and use Paste and translate. The original text is retained if an attempt fails, so you can retry. Without a configured text model, a translation will not appear automatically.

No sound, an inactive shortcut or failed text insertion

• No microphone level: check the selected device, system input volume and microphone permissions first.
• Shortcut not working: check for key conflicts and system keyboard-monitoring permissions. Text insertion on macOS may also require Accessibility permission.
• Linux / Wayland: shortcuts, text selection and insertion depend on your desktop’s portal or bridge support. Configure them for your actual environment.
• Recognition request failed: check the channel’s service type, URL, credentials and network connection. A remote speech API does not provide fully offline recognition.

A useful combination

Place the cursor in the AI Assistant input field, dictate the full task, and check the transcription before sending it. When reading material in another language, translate a selection first, then give the assistant a request you understand. Voice input reduces typing; reviewing the text before sending still helps prevent recognition errors from changing your intent.