Talk to it. Listen back.

What you'll get. Send a message with your voice instead of your thumbs, and have the answer read out loud while your hands are busy.

Start talking

There is no separate voice screen to find. You open a conversation the normal way and tap one button.

1. Open any chat. Frank, a provider agent, a paired runtime — it does not matter which.
2. Tap the round button at the right-hand end of the message box. A strip appears across the top of the chat. That strip is Conversation mode.
3. Say what you want. The strip narrates itself as it goes — Listening, Processing transcript, Speaking — so you always know which half of the turn you are in.
4. Stop by stopping. When you go quiet for your pause setting, the message sends on its own. Tapping the button again sends it immediately.

The pause is four seconds unless you have chosen otherwise. You can change it under Settings → Voice/Speech → Speech behavior, on the row called “Pause timeout” — “Silence before a voice message sends.” The choices are two, four and six seconds. If you picked two seconds at some point in the past, you still have two seconds; the four-second default only applies to people who never chose.

One button, three gestures

• Tap — start listening, or stop and send.
• Double-tap — pause the reply that is playing, and tap twice again to carry on.
• Long-press — mute the microphone. A red slash appears across the button, and it stays muted through the end of the run, through leaving the screen, and through a restart. Only you clear it, with another long press.

Having the answer read back

Under any assistant reply there is a Read out loud button. Tap it and that message plays. On a reply that is only an image, the same button reads Read image aloud and speaks the caption.

If you would rather never tap it, turn on “Speak final chat replies” in Settings → Voice/Speech → Speech behavior. Its own description is honest about the scope: “Read the completed assistant answer aloud after a voice turn.”

Tapping Read out loud claims the speaker for that one message. Nothing else can talk over it — not the agent's live progress narration, not the next reply — until your message has finished playing and the agent's turn has settled. Before that rule existed, working notes arriving mid-playback would start a second read in a different voice, on top of the first.

Interrupting it

You can simply talk. Words like stop, wait, pause, hold, cancel, no and actually are treated as a deliberate interruption and the playback yields to you. There is also a Stop action on the notification E:Voice leaves in your shade while a voice session is live, and an X on the Conversation mode strip itself.

Long answers, spoken sensibly

A long written reply is not a good spoken reply, so E:Voice speaks a shortened version by default and leaves the full text in the chat. The setting is “Speak a voice-optimized version”, with a Voice summary length slider: Concise, Balanced (the default), Detailed, or Full. Anything under 240 characters is spoken whole regardless.

Structure gets described rather than recited. A code block becomes “A code block with the install command.” A table becomes “A table with four rows and three columns.” A heading becomes “Section: …”. The complete, copyable version stays in the thread where you left it.

When the phone rings

An incoming call pauses the voice session, and the call is never fed to the transcriber as your speech. About two and a half seconds after the call ends, listening picks back up on its own. The same happens for a navigation prompt or any other app that briefly grabs the audio.