[App][Voice Chat] Add push-to-talk microphone input without ending the call

What variant of Codex are you using?

Codex App on Windows

What feature would you like to see?

Please add a true push-to-talk input mode specifically for realtime Voice Chat.

When this mode is selected, the Voice Chat session stays active and the user can continue hearing the AI, but the microphone is closed to the AI by default. Microphone audio is transmitted only while the user deliberately holds the configured push-to-talk key. Releasing the key immediately stops upstream microphone transmission without ending, pausing, or deactivating the Voice Chat session.

Pressing the key should be the explicit signal that the user is speaking to the AI. Speech while the key is not held must not be treated as AI input.

This is different from:

• a hotkey that starts or stops Voice Chat;
• a microphone toggle that remains on after one press;
• speech-to-text dictation in the text composer.

A configurable keyboard shortcut and a clear on-screen held/listening state would be helpful.

Additional information

Use case: realtime Voice Chat is active in a shared workplace. The user may be listening to the AI together with coworkers, while also talking with those coworkers about matters that are not addressed to the AI. Videos or other ambient audio may also be present. None of that background conversation or media audio should be uploaded to the AI.

Expected default behavior in push-to-talk mode:

1. Start realtime Voice Chat and keep the session active.
2. Continue hearing the AI while microphone transmission remains off by default.
3. Hold the push-to-talk key only when intentionally speaking to the AI.
4. Release the key to stop transmitting microphone audio immediately, while staying in the same Voice Chat session and continuing to hear the AI.

The key press creates an unambiguous boundary between speech directed to the AI and local conversation between people.

The issue names no files, tests, or entry points; begin by locating the Windows app's realtime Voice Chat and microphone-input implementation. Done means a configurable held key gates microphone transmission, release stops upload immediately, the session and AI audio remain active, and the UI shows the held/listening state.