Dictate a message

Purpose

Parisa has a long brief in her head and her hands full of sketches, so typing it would take longer than saying it. Speak your message and have it typed into the message box, ready to edit and send.

Before you start

Use a browser that can record from the microphone and allow microphone access when the browser asks. Dictation only types text for you. It is different from Talk using live voice, where the agent answers by voice. Dictation uses your usage allowance even if you do not send the resulting text.

Steps

In the message box, select the microphone button ("Dictate"). It sits next to the model button. The box switches to a recording bar with a moving sound wave and a timer.
Speak. To throw the recording away, select the red x ("Cancel dictation").
When you finish, select the check button ("Finish dictation"). While your words are turned into text, the check button becomes a spinner and the sound wave fades. A recording stops by itself after 2 hours and is then transcribed, so you can dictate the rest afterwards. In the last 5 minutes the bar shows "[time] left", and it turns red in the last 30 seconds.
The text appears at the end of what you had already typed, and the cursor waits at its end. Read it, fix any names or numbers, then send it with "Send message". Nothing is sent automatically, and the recording is neither attached to the message nor kept.
If the text could not be created, a message appears and the bar offers "Try again", which sends the same recording once more without recording again. Select "Cancel dictation" to give up and type instead.

Words Agentum learns

If you change a dictated word before sending (for example "a Gentoo" to "Agentum"), a small message appears at the bottom center of the screen: Adding "Agentum" to your dictionary, with a "Cancel" button. After 4 seconds the word is added and shows in Dictionary with a "Learned" label. Select "Cancel" and Agentum will not ask about that word again. Only respellings are learned, not rewording, and links, email addresses, file paths and numbers are skipped.

Expected result

Your spoken words are in the message box as normal text. The conversation is not touched until you press send.

Troubleshooting

Problem	What to do
"Dictation is not supported in this browser."	Update your browser or open Agentum in another current browser.
"Microphone access was not granted."	Allow the microphone for this site in your browser's address bar settings, then try again.
"No speech was detected."	Speak closer to the microphone, check that it is not muted, and record again.
"The voice message could not be transcribed. Please try again or type your message."	Select "Try again". If it keeps failing, type your message.
The spinner keeps turning for a long time	Wait, or select "Cancel dictation" and record a shorter message.
Transcription is blocked by your plan or usage limit	If your plan or usage limit blocks transcription, type the message instead or wait for the allowance to reset. See Understand plans and usage limits.
Names or numbers are wrong	Dictation can mishear them. Always read the text before sending. Add names and special words to Dictionary so they are spelled correctly next time.
Dictation is refused in a private workspace, project or chat	Dictation needs a private transcription model while Data Privacy Mode applies. Type the message instead. See Use Data Privacy Mode.