fix: infer actual system audio clock rate

2026-08-27 13:02:00 +02:00
parent 3a9ad8fe0c
commit 204857cdd9
3 changed files with 139 additions and 9 deletions
@@ -27,6 +27,7 @@ private struct SystemCaptureDiagnostics {
    var targetFormat = "unavailable"
    var selectedInputSampleRate: Double?
    var targetSampleRate: Double?
+    var inputRateInference = "unavailable"
    var firstBufferLayout: String?
    var callbackCount: UInt64 = 0
    var inputFrameCount: UInt64 = 0
@@ -443,18 +444,67 @@ final class AudioManager: NSObject, ObservableObject {
        systemDiagnostics.targetSampleRate = targetFormat.sampleRate
        var inputFormat: AVAudioFormat?
        var converter: AVAudioConverter?
+        var pendingBuffer: AVAudioPCMBuffer?
+        var pendingInputTime: AudioTimeStamp?
+        var pendingNow: AudioTimeStamp?
        try tap.run(on: tapQueue) { [weak self] inNow, inputData, inInputTime, _, _ in
            guard let self else { return }
            if inputFormat == nil {
+                guard let callbackFormat = self.inputFormat(
+                    for: inputData,
+                    sampleRate: advertisedInputFormat.sampleRate
+                ),
+                let currentBuffer = self.copyAudioBuffer(from: inputData, format: callbackFormat) else {
+                    self.systemDiagnostics.discardedCallbackCount += 1
+                    return
+                }
+
+                guard let previousBuffer = pendingBuffer,
+                      let previousInputTime = pendingInputTime else {
+                    pendingBuffer = currentBuffer
+                    pendingInputTime = inInputTime.pointee
+                    pendingNow = inNow.pointee
+                    return
+                }
+
+                let selectedSampleRate = self.effectiveInputSampleRate(
+                    advertisedSampleRate: advertisedInputFormat.sampleRate,
+                    previousFrameLength: previousBuffer.frameLength,
+                    previousTimestamp: previousInputTime,
+                    currentTimestamp: inInputTime.pointee
+                )
                inputFormat = self.inputFormat(
                    for: inputData,
-                    advertisedFormat: advertisedInputFormat
+                    sampleRate: selectedSampleRate
                )
                if let inputFormat {
                    self.systemDiagnostics.selectedInputFormat = self.audioFormatSummary(inputFormat)
                    self.systemDiagnostics.selectedInputSampleRate = inputFormat.sampleRate
                    converter = AVAudioConverter(from: inputFormat, to: targetFormat)
                }
+
+                guard let inputFormat, let converter else { return }
+                self.processAudioBuffer(
+                    { self.copyAudioBuffer(from: previousBuffer.audioBufferList, format: inputFormat) },
+                    converter: converter,
+                    targetFormat: targetFormat,
+                    source: .system,
+                    callbackTimestamp: previousInputTime,
+                    ioTimestamp: pendingNow
+                )
+                pendingBuffer = nil
+                pendingInputTime = nil
+                pendingNow = nil
+
+                self.processAudioBuffer(
+                    { self.copyAudioBuffer(from: currentBuffer.audioBufferList, format: inputFormat) },
+                    converter: converter,
+                    targetFormat: targetFormat,
+                    source: .system,
+                    callbackTimestamp: inInputTime.pointee,
+                    ioTimestamp: inNow.pointee
+                )
+                return
            }
            guard let inputFormat, let converter else { return }
            // The tap queue is serial. Reusing the converter preserves its
@@ -475,7 +525,7 @@ final class AudioManager: NSObject, ObservableObject {

    private func inputFormat(
        for inputData: UnsafePointer<AudioBufferList>,
-        advertisedFormat: AVAudioFormat
+        sampleRate: Double
    ) -> AVAudioFormat? {
        let buffers = UnsafeMutableAudioBufferListPointer(
            UnsafeMutablePointer(mutating: inputData)
@@ -492,12 +542,58 @@ final class AudioManager: NSObject, ObservableObject {
        let isInterleaved = buffers.count == 1 && channelCount > 1
        return AVAudioFormat(
            commonFormat: .pcmFormatFloat32,
-            sampleRate: advertisedFormat.sampleRate,
+            sampleRate: sampleRate,
            channels: AVAudioChannelCount(channelCount),
            interleaved: isInterleaved
        )
    }

+    private func effectiveInputSampleRate(
+        advertisedSampleRate: Double,
+        previousFrameLength: AVAudioFrameCount,
+        previousTimestamp: AudioTimeStamp,
+        currentTimestamp: AudioTimeStamp
+    ) -> Double {
+        let timestampsAreValid = previousTimestamp.mFlags.contains(.sampleTimeValid)
+            && currentTimestamp.mFlags.contains(.sampleTimeValid)
+        let sampleTimeDelta = currentTimestamp.mSampleTime - previousTimestamp.mSampleTime
+        guard timestampsAreValid,
+              advertisedSampleRate > 0,
+              previousFrameLength > 0,
+              sampleTimeDelta.isFinite,
+              sampleTimeDelta > 0 else {
+            systemDiagnostics.inputRateInference = "advertised (sample timestamps unavailable)"
+            return advertisedSampleRate
+        }
+
+        let inferredSampleRate = advertisedSampleRate * Double(previousFrameLength) / sampleTimeDelta
+        let plausibleRange = (advertisedSampleRate * 0.25)...(advertisedSampleRate * 1.25)
+        guard inferredSampleRate.isFinite, plausibleRange.contains(inferredSampleRate) else {
+            systemDiagnostics.inputRateInference = String(
+                format: "advertised (invalid inference %.3f from %u frames / %.3f sample-time units)",
+                inferredSampleRate,
+                previousFrameLength,
+                sampleTimeDelta
+            )
+            return advertisedSampleRate
+        }
+
+        let commonSampleRates: [Double] = [8_000, 11_025, 12_000, 16_000, 22_050, 24_000, 32_000, 44_100, 48_000, 88_200, 96_000]
+        let selectedSampleRate = commonSampleRates
+            .min(by: { abs($0 - inferredSampleRate) < abs($1 - inferredSampleRate) })
+            .flatMap { abs($0 - inferredSampleRate) / $0 <= 0.01 ? $0 : nil }
+            ?? inferredSampleRate
+        systemDiagnostics.inputRateInference = String(
+            format: "advertised=%.3f,inferred=%.3f,selected=%.3f,previousFrames=%u,sampleTimeDelta=%.3f",
+            advertisedSampleRate,
+            inferredSampleRate,
+            selectedSampleRate,
+            previousFrameLength,
+            sampleTimeDelta
+        )
+        return selectedSampleRate
+    }
+
    private func copyAudioBuffer(
        from inputData: UnsafePointer<AudioBufferList>,
        format: AVAudioFormat
@@ -863,6 +959,7 @@ final class AudioManager: NSObject, ObservableObject {
            "tapAdvertisedFormat=\(systemDiagnostics.tapAdvertisedFormat)",
            "aggregateInputFormat=\(systemDiagnostics.aggregateInputFormat)",
            "selectedInputFormat=\(systemDiagnostics.selectedInputFormat)",
+            "inputRateInference=\(systemDiagnostics.inputRateInference)",
            "targetFormat=\(systemDiagnostics.targetFormat)",
            "firstBufferLayout=\(systemDiagnostics.firstBufferLayout ?? "unavailable")",
            "callbackCount=\(systemDiagnostics.callbackCount)",