[macOS] Toy Audio Visualizer blocks Dictation and produces an empty recording

What version of the Codex App are you using?

ChatGPT/Codex Desktop 26.831.20005 (build 7524)

Platform

• macOS 26.6.2 (25G83)
• Apple Silicon Mac mini (M4), 16 GB RAM
• Default microphone: Jabra SPEAK 510 USB, mono, 16 kHz

What issue are you seeing?

When the Toy Audio Visualizer is enabled/open, Dictation in the Codex composer becomes very sluggish and does not capture audio. Stopping Dictation produces an empty recording/no transcription.

Disabling or closing the Toy Audio Visualizer restores Dictation.

During reproduction, the app launches the system-audio-spectrum helper. Codex Renderer, the Codex GPU service, and WindowServer also show elevated CPU usage while the visualizer is active.

Microphone permission was reset with:

tccutil reset Microphone com.openai.codex

The app was fully restarted and microphone access was granted again. The issue still reproduced while the visualizer was enabled, so it is not resolved by refreshing microphone permission.

No recording contents or private logs are attached.

Steps to reproduce

1. Open ChatGPT/Codex Desktop on macOS.
2. Enable/open the Toy Audio Visualizer.
3. Open a Codex task and click the Dictation microphone in the composer.
4. Speak, then stop Dictation.
5. Observe that the recording UI is sluggish and the resulting recording is empty/no transcription is produced.
6. Disable/close the Toy Audio Visualizer.
7. Try Dictation again; it works normally.

Expected behavior

The Toy Audio Visualizer should coexist with Dictation, or release/suspend its audio capture resources while Dictation is recording. Dictation should receive microphone frames and produce a transcription.

Actual behavior

With the Toy Audio Visualizer active, Dictation opens slowly but receives no usable audio frames and produces an empty recording.

Reproduce on macOS by enabling the Toy Audio Visualizer, then using Dictation in the Codex composer while watching the system-audio-spectrum helper. Trace how the visualizer and Dictation acquire microphone resources; done means Dictation receives audio and produces a transcription while the visualizer remains open or is suspended during recording.