Feature request: Supported live voice-caption stream for external displays

What variant of Codex are you using?

Codex app, voice conversations.

What feature would you like to see?

A supported, opt-in, read-only stream of the assistant text associated with an active voice response, emitted incrementally as the response is spoken rather than only after a completed transcript segment is saved.

This would enable captions on external displays and accessibility-oriented companion interfaces without requiring a separate voice session, retranscribing audio, or relying on internal log formats.

Useful properties:

• Incremental caption updates with stable response identifiers and ordering.
• Clear finalization, interruption, and correction events.
• Explicit user permission, session-scoped access, and easy revocation.
• No audio capture or access to unrelated conversations required.
• A documented local subscription or equivalent supported integration surface.

Additional information

The goal is to preserve the existing voice conversation while displaying its current spoken text elsewhere with minimal additional delay. Completed transcript segments can arrive too late for comfortable live captions. A supported caption feed would avoid fragile integrations with internal persistence behavior.

This is a general integration request, not a request for a specific hardware vendor or access to hidden reasoning.

No files or tests are named. Start by locating the voice-conversation response and transcript event entry points, then determine whether a supported local subscription surface exists. Done means an opt-in, revocable stream provides ordered incremental captions with stable identifiers and clear finalization, interruption, and correction events without exposing audio or unrelated conversations.