System audio pipeline and ScreenCaptureKit

Capturing clean call audio on macOS historically required intrusive third-party virtual audio drivers (such as Soundflower or BlackHole) or kernel extensions. Oyma eliminates this complexity by leveraging Apple’s modern native ScreenCaptureKit framework.

The ScreenCaptureKit architecture

ScreenCaptureKit was introduced by Apple to provide high-performance, sandboxed media capture. Beginning in macOS 14.2, Apple added system audio tap capabilities:

┌─────────────────┐        ┌──────────────────┐
│  Host Mic Input │        │ Call App Audio   │
│ (CoreAudio Tap) │        │(ScreenCaptureKit)│
└────────┬────────┘        └────────┬─────────┘
         │                          │
         ▼                          ▼
┌─────────────────────────────────────────────┐
│       Dual-Channel PCM Ring Buffer          │
│   (Channel 0: Mic, Channel 1: Loopback)     │
└──────────────────────┬──────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────┐
│       On-Device Whisper Speech Engine       │
└─────────────────────────────────────────────┘

Advantages of native ScreenCaptureKit

1. Zero System Modifications: No kernel extensions, audio server plugins, or administrator driver installers.
2. Apple Hardened Runtime: Fully sandboxed execution adhering strictly to Apple’s security policies.
3. Hardware Acceleration: Native CoreAudio buffer handling minimizes latency and preserves battery life.
4. App Isolation: Routes system audio directly from active conferencing apps without recording background system chimes.

Required macOS permissions

To capture both sides of a meeting, two macOS permissions are necessary:

• Microphone: Captures your spoken voice.
• Screen & System Audio Recording: Captures the other participants playing through your Mac speakers or headphones.

You can verify and test both permissions in Settings › Meetings › Diagnostics.

Related guides

• Permission setup in Recording permissions on macOS.
• Meeting app detection in Meeting app detection.
• Speech inference in Transcription models.