[Windows][26.814.5517.0] Realtime Voice audio works, but user and assistant transcripts are not rendered

What version of the Codex App are you using?

Codex Desktop for Windows 26.814.5517.0.

What subscription do you have?

Paid ChatGPT subscription (exact tier omitted).

What platform is your computer?

Windows 11 Pro 23H2, version 10.0.22631, x64. Locale and UI language: zh-CN.

What issue are you seeing?

Realtime Voice audio works normally in both directions, but the visible conversation does not render live transcript text for either the assistant or the user.

For example, I can hear the assistant say “Hello”, but no “Hello” text appears in the conversation. When I speak, my speech is recognized well enough for the voice conversation to continue, but my transcript is not displayed either. The voice orb/avatar remains visible and the audio conversation continues. Regular non-voice messages still render normally.

This is a regression for me: the same voice-practice workflow previously displayed both sides of the conversation in real time.

What steps can reproduce the bug?

1. Open an existing Codex task that has previously used Realtime Voice.
2. Start Realtime Voice in that task.
3. Let the assistant speak; for example, it says “Hello”.
4. Speak a reply.
5. Watch the visible conversation while both sides are speaking.
6. Audio continues normally, but neither side's transcript is rendered.

The issue reproduces in the affected existing task. I noticed it after returning to that task following about one week of inactivity. I have not confirmed whether every newly created task is affected.

What is the expected behavior?

• The assistant's spoken response should appear as text in real time.
• The user's recognized speech should appear as text in real time.
• The visible transcript should remain synchronized with the audio conversation.

Additional information

• Voice workflow: an IELTS speaking-practice loop in an existing local task.
• Local diagnostics show that the realtime conversation starts successfully with:
• model: gpt-live-1-codex
• output_modality: Audio
• codex_responses_as_items: false
• No transcript/caption toggle is exposed in the desktop app settings.
• No error banner appears in the UI.
• A potential diagnostic area is handling/rendering of thread/realtime/transcript/delta and thread/realtime/transcript/done events. This is a diagnostic hint, not a confirmed root cause.
• No raw logs, session transcripts, account details, thread IDs, or local filesystem paths are attached because they may contain private voice content. A screenshot and sanitized excerpts can be provided privately if requested.

Possibly related but not claimed as exact duplicates:

• #39178 — Windows build from the same release family; a hidden voice/avatar renderer can own a thread while the primary UI fails to render the correct state.
• #38965 — Realtime Voice can leave inconsistent/orphaned task UI state.
• #37900 — Voice-view transcript rendering can omit content on macOS.
• #37899 — Advanced Voice/live transcript UI can fail to initialize on an existing task.

This report was prepared with AI assistance from local diagnostic evidence and submitted with the user's approval.