Skip to main content
Use Voice to dictate a message or listen to an answer. Recording and playback are separate controls: you can use either without turning on the other. Open your profile menu, then go to Settings → Voice.
1

Turn on Voice mode

Voice mode places the microphone controls beside the message box. Turning it off hides those controls and stops an active recording.
2

Choose a microphone

Leave System default selected, or choose a specific input when your device has more than one.
3

Test before you record

Select Test and speak at a comfortable volume. The moving level bars confirm that the browser is receiving sound. Select Stop when you are finished; a test never sends a message.
4

Check the transcript

Names, dates, measurements, and medication doses deserve a second look. Correct the text before sending whenever a changed word could change the meaning.

​
Speaking and listening are separate

​
Speak to the Assistant

Voice mode records speech for transcription. You remain in control of the message and can review the words before they enter the conversation.

​
Listen to an answer

Read responses aloud plays new completed answers. It does not turn on the microphone. You can also use Read aloud on an individual response.

​
Know what the microphone is telling you

The browser has not tested the selected microphone in this session. Select Test when you are ready.
A successful test proves that the browser can hear the selected input. It does not prove that every word will be transcribed correctly. Quiet rooms, a steady distance from the microphone, and a quick review still matter.

​
Teach transcription the words that matter to you

The Transcription dictionary gives the speech system hints for names and vocabulary it may not expect. Add a person, medicine, organization, or specialist term exactly as you want it recognized.

​
Dictionary limits

Add up to 100 entries. Each entry can contain 50 characters, with 2,048 characters across the dictionary.

​
Duplicate entries

Empty entries and duplicates are ignored. Capitalization alone does not create a second entry.
Dictionary entries are suggestions, not replacements. They improve context for transcription but do not guarantee a spelling. Clearing the dictionary offers an undo action, so an accidental reset can be restored while that action is still available.

​
Keep only the transcripts you need

Recent transcripts are a working surface, not a second chat archive. The latest 20 completed transcripts are kept in encrypted, account-scoped storage for this browser. From the list, you can copy one transcript or delete it individually.

​
Clean up transcripts

Turn on Clean up transcripts when you want the app to reduce obvious transcription clutter. Keep it off when preserving the raw wording matters more. In either case, review health details before sending.
If encrypted local storage is unavailable, the Voice workspace reports that the settings cannot be loaded on this device. Changing browsers or clearing browser data can also remove this local history. Messages you already sent remain governed by the conversation and data controls described in Data, backups, and deletion.

​
Choose how answers sound

Uses the voices installed or exposed by your browser and device. Language filters that list; it does not translate an answer. Choosing another language clears the previous voice selection so you can choose a compatible voice.
These controls change capture and playback. They do not change the meaning, evidence, or safety limits of an answer.

​
Microphone access and privacy

Your browser asks for microphone permission. You can change that permission in its site settings. Testing the microphone does not create a chat message. Transcription processes your recorded speech; cloud read-aloud processes the answer text needed to produce audio. Do not paste a transcript into public feedback. For private account help, use Contact us and leave out passwords, authentication codes, recovery material, and anything the support team does not need.

​
Questions and answers

Test the selected microphone in Settings → Voice. If it says Permission blocked, allow access in the browser’s site settings. If it says No microphone found, reconnect the input or choose another one.
No. Read-aloud controls playback of completed answers. Voice mode controls microphone recording and transcription.
No. Dictionary entries are recognition hints. Review the transcript, especially when a name, number, measurement, or dose affects the meaning.
The latest 20 are kept in encrypted, account-scoped storage for the current browser. They are not a substitute for your conversation archive and may disappear when local browser data is cleared.
No. A test confirms that the browser can hear the selected input. It does not create or send a message.