Talon/voice-cat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs: condense implementation comments
Some checks failed
Build Linux Binaries / linux/amd64 (push) Has been cancelled
Details
Build Linux Binaries / linux/arm64 (push) Has been cancelled
Details

Browse Source
This commit is contained in:
Talon
2026-07-23 13:37:05 +02:00
parent 575e2907d0
commit 4f71b784fe
22 changed files with 102 additions and 507 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
clients/apple/macOS/VoiceCatMac/Audio/ScreenAudioCapture.swift
View File

@@ -1,14 +1,11 @@
import AVFoundation
import ScreenCaptureKit
// Which apps' audio the SCREEN_AUDIO stream captures. ScreenCaptureKit filters audio at the
// *application* level (not per-window), so the selection is expressed as bundle IDs. The
// picker UI (ScreenSharePickerSheet) produces a `ScreenAudioSelection`; `start()` turns it
// into the matching `SCContentFilter`.
// ScreenCaptureKit filters audio by application bundle identifier.
enum ScreenAudioScope: Equatable {
case entireDesktop // whole display — the original behaviour
case onlyApps([String]) // capture only these bundle IDs
case allExcept([String]) // capture everything except these bundle IDs
case entireDesktop
case onlyApps([String])
case allExcept([String])
}
struct ScreenAudioSelection: Equatable {
@@ -20,20 +17,8 @@ struct ScreenAudioSelection: Equatable {
static let `default` = ScreenAudioSelection()
}
// ScreenAudioCapture — macOS system/desktop audio capture for the SCREEN_AUDIO stream.
//
// The macOS analog of the Windows WASAPI loopback path (docs/voice.md §9). ScreenCaptureKit
// (macOS 13+) captures whatever the system is playing; we convert each audio CMSampleBuffer
// (Float32) → int16 interleaved and push 20 ms frames (960 samples/channel @ 48 kHz) into the
// core via `vc_stream_feed_pcm` (exposed as `VoiceCatClient.feedPcm`). The core then runs the
// same Opus-encode → media-AEAD → UDP path as any other stream — only the *source* is
// platform-specific (architecture.md §4).
//
// Audio-only: we request a 2×2 video plane at 1 fps purely because SCStream needs a video
// configuration, and we never add a `.screen` output — only `.audio`. `excludesCurrentProcess
// Audio` prevents the self-echo loop of re-capturing our own incoming voice mix.
//
// `feedPcm` is thread-safe (any thread), so we forward straight from the sample-handler queue.
// SCStream requires a minimal video configuration even for audio-only capture. Only its audio
// output is registered, and current-process audio is excluded to prevent feedback.
final class ScreenAudioCapture: NSObject, SCStreamOutput, SCStreamDelegate {
/// Receives a full 20 ms frame: (interleaved int16 PCM, samplesPerChannel = 960, channels).