[Windows][Codex app] AirPods Pro 3 input repeatedly switches to microphone array during active dictation
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.901.6511.0 x64
What subscription do you have?
ChatGPT Plus
What platform is your computer?
Microsoft Windows NT 10.0.22631.0 x64
What issue are you seeing?
Composer dictation initially starts correctly with AirPods Pro 3 and works for a short time. During the active dictation session, Windows Sound settings then begins repeatedly switching the active input between:
- the AirPods Pro 3 microphone
- the laptop's built-in digital microphone array
Dictation consequently alternates between working and not working. The AirPods also produce a connection/reconnection-like signal during the switching.
The problem appears specific to composer dictation in the Codex desktop app:
- Continuous Codex voice chat works normally with the same AirPods.
- ChatGPT in the web browser works normally with the same AirPods.
- If another application is already playing audio before dictation starts, the repeated input switching does not occur. The background audio becomes muffled or lower-quality while the microphone is active, but dictation remains stable.
This suggests the Codex desktop dictation session loses the AirPods capture endpoint after recording has already started, causing Windows to repeatedly fall back to the built-in microphone array and then return to the AirPods.
What steps can reproduce the bug?
- Connect AirPods Pro 3 to Windows.
- Ensure no other application is playing audio.
- Select the AirPods microphone as the input device.
- Open the Codex desktop app.
- Start composer dictation using the microphone button in the prompt input.
- Speak normally. Dictation initially works.
- After a short time, observe the active input in Windows Sound settings.
- The input repeatedly switches between the AirPods Pro 3 microphone and the built-in digital microphone array.
- Dictation alternates between working and not working.
- Start continuous low-volume audio in another application and repeat the test.
- The endpoint switching no longer occurs; the background audio only becomes lower-quality while the microphone is active.
What is the expected behavior?
Once composer dictation successfully starts with the selected AirPods microphone, Codex should keep that recording endpoint active until the user stops dictation or a clear device error occurs.
Additional information
- Continuous Codex voice chat is unaffected.
- ChatGPT's web app is unaffected.
- Keeping another playback stream active prevented the problem in the observed tests.
- The AirPods microphone therefore works with other capture paths on the same computer.
- Related but distinct: #39409 concerns an established AirPods microphone session stopping on macOS. This report concerns repeated Windows input-endpoint switching during an active dictation session.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing composer dictation in the Codex desktop app on Windows with AirPods Pro 3 and no background audio, then compare it with continuous voice chat and browser ChatGPT. Trace the microphone-button recording path and observe Windows Sound input changes during an active session. Done means the selected AirPods recording endpoint remains active until dictation stops or a clear device error occurs.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- audio-video-rtc, desktop, operating-systems
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 48/100