Voice typing and read aloud
Dictate a draft through the composer and listen to assistant replies with the speech settings that fit the account.
Speech across surfaces
Speech settings belong to the AI Integrator account. Chat supports the dictation and read-aloud controls described below where the application exposes them. Desktop Code also offers transcript read aloud when available, using the account speech service rather than a coding runtime's voice capability. The selected voice and speech model do not change the task's coding model. A phone's Remote connection does not provide a second local speech or coding runtime.
Dictate into the composer
The microphone button beside the composer starts voice typing. Its labels move from Start voice typing to Stop voice recording while the capture is active. The recording HUD shows a timer and the instruction Stop to transcribe · tap composer or Esc to cancel. Stopping sends the clip for transcription and appends the returned text to the existing draft; the normal send control then sends that draft as a chat message.
Voice capture ends automatically at 30 seconds. Cancel discards the recording and leaves the draft unchanged. A browser or device microphone permission is required, and an error appears in the composer when capture or transcription cannot start. The composer can still contain typed text before dictation begins, which makes a short spoken addition practical.
Read a reply aloud
Assistant messages have a Read aloud action. The control reports when speech is being generated, when playback is active, and when saved audio can be played again. The voice and text-to-speech model are selected in Settings > Speech & Voice. That page also selects the Speech to text model and lists the voices supported by the chosen text-to-speech model.
Speech clips are sent to the selected speech service when the action is used, and speech usage can be billed under the account’s plan. The settings page presents the applicable privacy and training choices alongside the model controls, so a voice configuration can be reviewed before recording or playback.
Generated read-aloud audio is saved under Files > Generated > Audio for a normal chat, where it can be previewed or downloaded. Private or incognito conversations do not add that audio to the library. See Files for library organization and Chat for model and message actions.