ARTICLE AD BOX
I'm building ScamAlert Pro, an Android app that detects scam calls in real time using on-device AI (Moonshine STT). The concept is proven — transcription, keyword detection, and overlay alerts all work perfectly in demo mode. The single blocker is mic access during live calls.
What we've tested and confirmed silenced (via adb shell dumpsys audio):
AudioSource.MIC — silenced
AudioSource.VOICE_COMMUNICATION — silenced
AudioSource.REMOTE_SUBMIX — fails to initialise, falls back to MIC, silenced
The audio dump shows this consistently:
rec update uid:10423 src:MIC silenced pack:com.scamalertpro.app
rec update uid:10423 src:VOICE_COMMUNICATION silenced pack:com.scamalertpro.app
Samsung Developer Program confirmed: "There are no APIs other than what Google Android provides officially for tapping mic data."
Our working theory: The telephony stack owns the voice session exclusively. Any third-party AudioRecord instance is silenced by the HAL the moment MODE_IN_CALL is active, regardless of audio source. The only path to unsuppressed audio capture is being the session owner.
SAP is already registered as the default dialer — but default dialer status alone doesn't appear to grant voice session ownership on Samsung devices.
Specific questions:
Does any AudioSource escape HAL silencing during MODE_IN_CALL on non-rooted Android 12+?
Does default dialer status grant voice session ownership on any OEM implementation?
Has anyone successfully implemented a virtual audio device or HAL shim that intercepts telephony audio without root?
Is there a documented path to CAPTURE_AUDIO_OUTPUT outside of being a system app?
This is for genuine user protection — not recording or storing audio. All processing is on-device, real-time, discarded immediately. Happy to share full audio dumps, code, or architecture details.
Any experience with this specific problem welcome — especially from call recorder developers who've navigated Samsung/Pixel HAL differences.
