bug: transcription stops when macOS system volume is set to zero

[devin-ai-integration]

Description

A user reported that transcription stops working when the macOS system volume is set to zero. It works correctly when volume is at 1 or above.

Reproduction: Set the Mac speaker volume to zero while a transcription session is active. Transcription stops producing output.

Analysis

The macOS speaker capture uses CAProcessTap (via cidre) in crates/audio/src/speaker/macos.rs. It creates a global audio tap via TapDesc::with_mono_global_tap_excluding_processes that should capture process audio before system volume is applied.

The app already plays a silent audio stream (AudioOutput::silence() in crates/audio/src/lib.rs) to keep the audio device active, stored as _silence_stream_tx in the SourceState.

Despite this, transcription stops at volume=0. The likely cause is a macOS-level power/processing optimization that stops the tap from receiving data when output volume is zero, even though the API is documented to capture pre-volume audio.

The VadMask (crates/vad-ext/src/masking.rs) only applies to the mic channel, not the speaker channel, so that is not the cause.

Potential Fixes

• Set tap_auto_start to true (currently false in macos.rs:69) — might help keep the tap active when no audio is flowing
• Detect when system volume is 0 and show a warning/notification to the user
• Investigate ScreenCaptureKit as an alternative capture path that is volume-independent

Relevant Files

• crates/audio/src/speaker/macos.rs — macOS speaker capture via CAProcessTap
• crates/audio/src/lib.rs — AudioOutput::silence() workaround
• plugins/listener/src/actors/source/mod.rs — _silence_stream_tx usage
• plugins/listener/src/actors/source/stream.rs — speaker stream setup