fix: count Core Audio callback frames explicitly

Meetingnotes.xcodeproj/project.pbxproj

@@ -276,7 +276,7 @@
 				CODE_SIGN_IDENTITY = "Apple Development";
 				CODE_SIGN_STYLE = Automatic;
 				COMBINE_HIDPI_IMAGES = YES;
-				CURRENT_PROJECT_VERSION = 38;
+				CURRENT_PROJECT_VERSION = 40;
 				DEVELOPMENT_ASSET_PATHS = "\"meetingnotes/Preview Content\"";
 				DEVELOPMENT_TEAM = G9LVHZAJNX;
 				ENABLE_HARDENED_RUNTIME = YES;
@@ -290,7 +290,7 @@
 					"@executable_path/../Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 15.0;
-				MARKETING_VERSION = 1.1.26;
+				MARKETING_VERSION = 1.1.28;
 				ONLY_ACTIVE_ARCH = NO;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;
@@ -312,7 +312,7 @@
 				CODE_SIGN_IDENTITY = "Apple Development";
 				CODE_SIGN_STYLE = Automatic;
 				COMBINE_HIDPI_IMAGES = YES;
-				CURRENT_PROJECT_VERSION = 38;
+				CURRENT_PROJECT_VERSION = 40;
 				DEVELOPMENT_ASSET_PATHS = "\"meetingnotes/Preview Content\"";
 				DEVELOPMENT_TEAM = G9LVHZAJNX;
 				ENABLE_HARDENED_RUNTIME = YES;
@@ -326,7 +326,7 @@
 					"@executable_path/../Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 15.0;
-				MARKETING_VERSION = 1.1.26;
+				MARKETING_VERSION = 1.1.28;
 				ONLY_ACTIVE_ARCH = YES;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;

meetingnotes/Managers/AudioManager.swift

@@ -443,14 +443,15 @@ final class AudioManager: NSObject, ObservableObject {
         let channelCount = buffers.reduce(UInt32(0)) { $0 + $1.mNumberChannels }
         guard channelCount > 0 else { return nil }
 
-        // HAL tap metadata can advertise interleaved stereo while the callback
-        // supplies one mono buffer per channel (or the reverse). Constructing a
-        // PCM buffer with that mismatched layout halves its frame count and
-        // produces 2x-speed system audio. The callback's AudioBufferList is the
-        // authoritative layout for the memory we are copying.
+        // HAL I/O proc samples use the canonical Float32 representation. The
+        // tap's stream description can advertise a different common format;
+        // using that to interpret the callback bytes can halve the frame count
+        // (for example, treating four-byte Float32 samples as eight-byte
+        // Float64 samples). The callback's AudioBufferList is authoritative for
+        // its channel layout, while its sample rate comes from the input stream.
         let isInterleaved = buffers.count == 1 && channelCount > 1
         return AVAudioFormat(
-            commonFormat: advertisedFormat.commonFormat,
+            commonFormat: .pcmFormatFloat32,
             sampleRate: advertisedFormat.sampleRate,
             channels: AVAudioChannelCount(channelCount),
             interleaved: isInterleaved
@@ -461,20 +462,27 @@ final class AudioManager: NSObject, ObservableObject {
         from inputData: UnsafePointer<AudioBufferList>,
         format: AVAudioFormat
     ) -> AVAudioPCMBuffer? {
-        guard let borrowedBuffer = AVAudioPCMBuffer(
-            pcmFormat: format,
-            bufferListNoCopy: inputData,
-            deallocator: nil
-        ), borrowedBuffer.frameLength > 0,
-        let ownedBuffer = AVAudioPCMBuffer(
-            pcmFormat: format,
-            frameCapacity: borrowedBuffer.frameLength
-        ) else { return nil }
-
-        ownedBuffer.frameLength = borrowedBuffer.frameLength
         let sourceBuffers = UnsafeMutableAudioBufferListPointer(
             UnsafeMutablePointer(mutating: inputData)
         )
+        let frameLengths = sourceBuffers.compactMap { source -> AVAudioFrameCount? in
+            let bytesPerFrame = Int(source.mNumberChannels) * MemoryLayout<Float32>.size
+            guard bytesPerFrame > 0,
+                  Int(source.mDataByteSize).isMultiple(of: bytesPerFrame) else { return nil }
+            return AVAudioFrameCount(Int(source.mDataByteSize) / bytesPerFrame)
+        }
+        guard frameLengths.count == sourceBuffers.count,
+              let frameLength = frameLengths.first,
+              frameLength > 0,
+              frameLengths.allSatisfy({ $0 == frameLength }),
+              let ownedBuffer = AVAudioPCMBuffer(
+                pcmFormat: format,
+                frameCapacity: frameLength
+              ) else { return nil }
+
+        // Set the frame count explicitly instead of asking AVAudioPCMBuffer to
+        // infer it from potentially inconsistent tap metadata.
+        ownedBuffer.frameLength = frameLength
         let destinationBuffers = UnsafeMutableAudioBufferListPointer(
             ownedBuffer.mutableAudioBufferList
         )

meetingnotes/ProcessTap/ProcessTap.swift

@@ -138,11 +138,11 @@ final class ProcessTap {
             tapDescription = CATapDescription(stereoMixdownOfProcesses: [process.objectID])
             logger.debug("Configuring tap for single process objectID: \(process.objectID)")
         case .systemAudio:
-            // Keep the HAL tap's buffer layout consistent with the default
-            // output stream. AudioManager performs the stereo-to-mono mix when
-            // it converts the captured audio to the 16 kHz transcription file.
-            tapDescription = CATapDescription(stereoGlobalTapButExcludeProcesses: [])
-            logger.debug("Configuring a stereo global system audio tap.")
+            // The transcription file is mono, so ask Core Audio for a mono
+            // mixdown at the source. This avoids interpreting a stereo HAL
+            // buffer as half as many frames before the 16 kHz conversion.
+            tapDescription = CATapDescription(monoGlobalTapButExcludeProcesses: [])
+            logger.info("Configuring a mono global system audio tap.")
         }
         
         tapDescription.uuid = UUID()