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 = 35;
+				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 = 1.1.23;
+				MARKETING_VERSION = 1.1.28;
 				ONLY_ACTIVE_ARCH = NO;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;
@@ -312,7 +312,7 @@
 				CODE_SIGN_IDENTITY = "Apple Development";
 				CODE_SIGN_STYLE = Automatic;
 				COMBINE_HIDPI_IMAGES = YES;
-				CURRENT_PROJECT_VERSION = 35;
+				CURRENT_PROJECT_VERSION = 40;
 				DEVELOPMENT_ASSET_PATHS = "\"meetingnotes/Preview Content\"";
 				DEVELOPMENT_TEAM = G9LVHZAJNX;
 				ENABLE_HARDENED_RUNTIME = YES;
@@ -326,7 +326,7 @@
 					"@executable_path/../Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 15.0;
-				MARKETING_VERSION = 1.1.23;
+				MARKETING_VERSION = 1.1.28;
 				ONLY_ACTIVE_ARCH = YES;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;

README.md

@@ -22,7 +22,8 @@ Implemented:
 - Meeting search functionality
 - Abilty to edit system prompt
 - Select any compatible Coder model for transcription and note generation
-- Automatic start and stop from MuteDeck through a compatible local API
+- Automatic start and stop from MuteDeck through a compatible local API, with a 10-second reconnect grace period
+- Merge an automatically split continuation back into its previous meeting
 - Auto updates
 - Text formatting
 - Different note templates

meetingnotes/Managers/AudioManager.swift

@@ -133,17 +133,29 @@ final class AudioManager: NSObject, ObservableObject {
         isProcessing = true
         defer { isProcessing = false }
         let model = UserDefaultsManager.shared.transcriptionModel
-        let micURL = recoveryFiles.first(where: { $0.source == .mic })?.url
+        let sessions = Dictionary(grouping: recoveryFiles) { recoverySessionKey(for: $0.url) }
-        let systemURL = recoveryFiles.first(where: { $0.source == .system })?.url
+            .values
+            .map { files in
+                (files: files, startedAt: recoveryCaptureStartedAt(for: files, fallback: captureStartedAt))
+            }
+            .sorted { $0.startedAt < $1.startedAt }
+
+        var chunks: [TranscriptChunk] = []
+        var failures: [String] = []
+        for session in sessions {
+            let micURL = session.files.first(where: { $0.source == .mic })?.url
+            let systemURL = session.files.first(where: { $0.source == .system })?.url
             async let micResult = transcribe(micURL, model: model, diarization: false)
             async let systemResult = transcribe(systemURL, model: model, diarization: true)
             let (micTranscription, systemTranscription) = await (micResult, systemResult)
-        let results = [micTranscription, systemTranscription]
+            let result = buildTranscriptChunks(
-        let (chunks, failures) = buildTranscriptChunks(
+                from: [micTranscription, systemTranscription],
-            from: results,
+                captureStartedAt: session.startedAt,
-            captureStartedAt: captureStartedAt,
+                existingChunks: chunks
-            existingChunks: []
             )
+            chunks = result.0
+            failures.append(contentsOf: result.1)
+        }
 
         if !failures.isEmpty {
             throw RecoveryTranscriptionError.requestFailed(failures.joined(separator: "; "))
@@ -154,6 +166,28 @@ final class AudioManager: NSObject, ObservableObject {
         return chunks
     }
 
+    private func recoverySessionKey(for url: URL) -> String {
+        let name = url.deletingPathExtension().lastPathComponent
+        if name.hasSuffix("-mic") { return String(name.dropLast(4)) }
+        if name.hasSuffix("-system") { return String(name.dropLast(7)) }
+        return name
+    }
+
+    private func recoveryCaptureStartedAt(
+        for files: [(url: URL, source: AudioSource)],
+        fallback: Date
+    ) -> Date {
+        let estimatedStarts = files.compactMap { file -> Date? in
+            guard let duration = audioDuration(at: file.url),
+                  let values = try? file.url.resourceValues(forKeys: [.contentModificationDateKey, .creationDateKey]),
+                  let finishedAt = values.contentModificationDate ?? values.creationDate else {
+                return nil
+            }
+            return finishedAt.addingTimeInterval(-duration)
+        }
+        return estimatedStarts.min() ?? fallback
+    }
+
     func cancelRecording() {
         cancelCapture(removeFiles: true)
         lastRecoveryAudioFolderName = nil
@@ -409,14 +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
+        // HAL I/O proc samples use the canonical Float32 representation. The
-        // supplies one mono buffer per channel (or the reverse). Constructing a
+        // tap's stream description can advertise a different common format;
-        // PCM buffer with that mismatched layout halves its frame count and
+        // using that to interpret the callback bytes can halve the frame count
-        // produces 2x-speed system audio. The callback's AudioBufferList is the
+        // (for example, treating four-byte Float32 samples as eight-byte
-        // authoritative layout for the memory we are copying.
+        // 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
@@ -427,20 +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/Managers/LocalStorageManager.swift

@@ -152,6 +152,98 @@ class LocalStorageManager {
         }
     }
 
+    func mergeMeeting(_ continuation: Meeting, into previous: Meeting) -> Meeting? {
+        guard continuation.id != previous.id, continuation.date >= previous.date else { return nil }
+
+        let previousFolder = recoveryDirectory.appendingPathComponent(previous.id.uuidString, isDirectory: true)
+        let continuationFolder = recoveryAudioFolder(named: continuation.id.uuidString)
+        var copiedAudioURLs: [URL] = []
+        var createdPreviousFolder = false
+
+        if let continuationFolder {
+            if !FileManager.default.fileExists(atPath: previousFolder.path) {
+                do {
+                    try FileManager.default.createDirectory(at: previousFolder, withIntermediateDirectories: true)
+                    createdPreviousFolder = true
+                } catch {
+                    return nil
+                }
+            }
+
+            for audioFile in recoveryAudioFiles(in: continuationFolder) {
+                let destination = previousFolder.appendingPathComponent(audioFile.url.lastPathComponent)
+                if FileManager.default.fileExists(atPath: destination.path) {
+                    guard FileManager.default.contentsEqual(
+                        atPath: audioFile.url.path,
+                        andPath: destination.path
+                    ) else {
+                        rollbackMergedAudio(copiedAudioURLs, removeFolder: createdPreviousFolder ? previousFolder : nil)
+                        return nil
+                    }
+                    continue
+                }
+
+                do {
+                    try FileManager.default.copyItem(at: audioFile.url, to: destination)
+                    copiedAudioURLs.append(destination)
+                } catch {
+                    rollbackMergedAudio(copiedAudioURLs, removeFolder: createdPreviousFolder ? previousFolder : nil)
+                    return nil
+                }
+            }
+        }
+
+        var merged = previous
+        var seenChunkIDs = Set<UUID>()
+        merged.transcriptChunks = (previous.transcriptChunks + continuation.transcriptChunks)
+            .filter { seenChunkIDs.insert($0.id).inserted }
+            .sorted {
+                if $0.timestamp == $1.timestamp {
+                    return $0.id.uuidString < $1.id.uuidString
+                }
+                return $0.timestamp < $1.timestamp
+            }
+        merged.userNotes = mergedText(previous.userNotes, continuation.userNotes)
+        merged.generatedNotes = mergedText(previous.generatedNotes, continuation.generatedNotes)
+        merged.templateId = previous.templateId ?? continuation.templateId
+        if !recoveryAudioFiles(in: previousFolder).isEmpty {
+            merged.recoveryAudioFolderName = previous.id.uuidString
+        }
+
+        guard saveMeeting(merged) else {
+            rollbackMergedAudio(copiedAudioURLs, removeFolder: createdPreviousFolder ? previousFolder : nil)
+            return nil
+        }
+
+        guard deleteMeeting(continuation) else {
+            _ = saveMeeting(previous)
+            rollbackMergedAudio(copiedAudioURLs, removeFolder: createdPreviousFolder ? previousFolder : nil)
+            return nil
+        }
+
+        if let continuationFolder {
+            try? FileManager.default.removeItem(at: continuationFolder)
+        }
+        return merged
+    }
+
+    private func mergedText(_ first: String, _ second: String) -> String {
+        let first = first.trimmingCharacters(in: .whitespacesAndNewlines)
+        let second = second.trimmingCharacters(in: .whitespacesAndNewlines)
+        guard !first.isEmpty else { return second }
+        guard !second.isEmpty, second != first else { return first }
+        return first + "\n\n---\n\n" + second
+    }
+
+    private func rollbackMergedAudio(_ copiedURLs: [URL], removeFolder folder: URL?) {
+        for url in copiedURLs {
+            try? FileManager.default.removeItem(at: url)
+        }
+        if let folder, recoveryAudioFiles(in: folder).isEmpty {
+            try? FileManager.default.removeItem(at: folder)
+        }
+    }
+
     // MARK: - Recovery Audio
 
     func preserveAudioFiles(_ urls: [URL], for meetingID: UUID) -> URL? {

meetingnotes/ProcessTap/CoreAudioUtils.swift

@@ -116,6 +116,47 @@ extension AudioObjectID {
         try read(kAudioTapPropertyFormat, defaultValue: AudioStreamBasicDescription())
     }
 
+    /// Reads the virtual format of the first input stream exposed by this device.
+    ///
+    /// An aggregate device can adapt a tap to the active output hardware. Its
+    /// input stream format is therefore the format delivered to the I/O proc,
+    /// which can differ from the tap object's originally advertised format.
+    func readInputStreamBasicDescription() throws -> AudioStreamBasicDescription {
+        var streamsAddress = AudioObjectPropertyAddress(
+            mSelector: kAudioDevicePropertyStreams,
+            mScope: kAudioObjectPropertyScopeGlobal,
+            mElement: kAudioObjectPropertyElementMain
+        )
+        var dataSize: UInt32 = 0
+        var status = AudioObjectGetPropertyDataSize(self, &streamsAddress, 0, nil, &dataSize)
+        guard status == noErr else {
+            throw "Error reading device stream list size: \(status)"
+        }
+
+        var streamIDs = [AudioObjectID](
+            repeating: .unknown,
+            count: Int(dataSize) / MemoryLayout<AudioObjectID>.size
+        )
+        status = AudioObjectGetPropertyData(self, &streamsAddress, 0, nil, &dataSize, &streamIDs)
+        guard status == noErr else {
+            throw "Error reading device stream list: \(status)"
+        }
+
+        for streamID in streamIDs {
+            let direction: UInt32 = try streamID.read(
+                kAudioStreamPropertyDirection,
+                defaultValue: 0
+            )
+            guard direction == 1 else { continue }
+            return try streamID.read(
+                kAudioStreamPropertyVirtualFormat,
+                defaultValue: AudioStreamBasicDescription()
+            )
+        }
+
+        throw "Device has no input stream."
+    }
+
     private func requireSystemObject() throws {
         if self != .system { throw "Only supported for the system object." }
     }

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
+            // The transcription file is mono, so ask Core Audio for a mono
-            // output stream. AudioManager performs the stereo-to-mono mix when
+            // mixdown at the source. This avoids interpreting a stereo HAL
-            // it converts the captured audio to the 16 kHz transcription file.
+            // buffer as half as many frames before the 16 kHz conversion.
-            tapDescription = CATapDescription(stereoGlobalTapButExcludeProcesses: [])
+            tapDescription = CATapDescription(monoGlobalTapButExcludeProcesses: [])
-            logger.debug("Configuring a stereo global system audio tap.")
+            logger.info("Configuring a mono global system audio tap.")
         }
         
         tapDescription.uuid = UUID()
@@ -248,8 +248,20 @@ final class ProcessTap {
 
         do {
             logger.debug("Attempting to read audio tap stream basic description for tapID #\(tapID)...")
-            self.tapStreamDescription = try tapID.readAudioTapStreamBasicDescription()
+            let advertisedDescription = try tapID.readAudioTapStreamBasicDescription()
-            logger.debug("Successfully read tap stream description: \(String(describing: self.tapStreamDescription))")
+
+            // The aggregate device may adapt the tap to the active hardware's
+            // sample rate. Its input stream is the format actually delivered
+            // to the I/O proc, so use that rather than the tap's pre-aggregate
+            // advertisement. Using the latter can halve the written duration
+            // when, for example, a 48 kHz tap is delivered at 24 kHz.
+            do {
+                self.tapStreamDescription = try aggregateDeviceID.readInputStreamBasicDescription()
+                logger.info("Using aggregate input stream description: \(String(describing: self.tapStreamDescription), privacy: .public); tap advertised: \(String(describing: advertisedDescription), privacy: .public)")
+            } catch {
+                self.tapStreamDescription = advertisedDescription
+                logger.warning("Could not read aggregate input stream format; using tap format: \(error, privacy: .public)")
+            }
         } catch {
             logger.error("Failed to read audio tap stream basic description for tapID #\(tapID): \(error)")
             throw error // Propagate error

meetingnotes/Services/LocalAPIServer.swift

@@ -200,7 +200,9 @@ private final class LocalRecordingController {
     static let shared = LocalRecordingController()
 
     private let recordingManager = RecordingSessionManager.shared
+    private let stopGraceNanoseconds: UInt64 = 10_000_000_000
     private var isStopping = false
+    private var pendingStopTask: Task<Void, Never>?
 
     private init() {}
 
@@ -237,6 +239,12 @@ private final class LocalRecordingController {
     }
 
     func startRecording() throws -> [String: Any] {
+        if let pendingStopTask {
+            pendingStopTask.cancel()
+            self.pendingStopTask = nil
+            isStopping = false
+            return statusPayload()
+        }
         if isStopping || recordingManager.isProcessing {
             throw LocalRecordingError.processing
         }
@@ -258,14 +266,22 @@ private final class LocalRecordingController {
             return statusPayload()
         }
         isStopping = true
-        Task { [weak self] in
+        pendingStopTask = Task { [weak self] in
+            try? await Task.sleep(nanoseconds: self?.stopGraceNanoseconds ?? 0)
+            guard !Task.isCancelled else { return }
             await self?.finishRecording()
         }
         return statusPayload()
     }
 
     func cancelRecording() -> [String: Any] {
-        guard !isStopping else { return statusPayload() }
+        if let pendingStopTask {
+            pendingStopTask.cancel()
+            self.pendingStopTask = nil
+            isStopping = false
+        } else if isStopping {
+            return statusPayload()
+        }
         let meetingID = recordingManager.activeMeetingId
         recordingManager.cancelRecording()
         if let meetingID,
@@ -277,6 +293,7 @@ private final class LocalRecordingController {
     }
 
     private func finishRecording() async {
+        pendingStopTask = nil
         let meetingID = recordingManager.activeMeetingId
         let chunks = await recordingManager.stopRecording()
         guard let meetingID,

meetingnotes/ViewModels/MeetingListViewModel.swift

@@ -67,6 +67,28 @@ class MeetingListViewModel: ObservableObject {
         _ = LocalStorageManager.shared.deleteMeeting(meeting)
     }
 
+    func previousMeeting(for meeting: Meeting) -> Meeting? {
+        meetings
+            .filter { $0.id != meeting.id && $0.date < meeting.date }
+            .max { $0.date < $1.date }
+    }
+
+    func mergeIntoPrevious(_ meeting: Meeting) -> Meeting? {
+        guard let previous = previousMeeting(for: meeting),
+              let merged = LocalStorageManager.shared.mergeMeeting(meeting, into: previous) else {
+            errorMessage = "The meetings could not be merged. Their original records and audio were kept."
+            return nil
+        }
+
+        meetings.removeAll { $0.id == meeting.id || $0.id == previous.id }
+        meetings.append(merged)
+        meetings.sort { $0.date > $1.date }
+        NotificationCenter.default.post(name: .meetingSaved, object: merged)
+        NotificationCenter.default.post(name: .meetingDeleted, object: meeting)
+        PostHogSDK.shared.capture("meetings_merged")
+        return merged
+    }
+    
     func createNewMeeting() -> Meeting {
         let newMeeting = Meeting(templateId: LocalStorageManager.shared.preferredTemplateID())
         meetings.insert(newMeeting, at: 0)

meetingnotes/Views/MeetingListView.swift

@@ -83,10 +83,19 @@ struct MeetingListView: View {
         NavigationStack(path: $navigationPath) {
             Group {
                 if let selectedMeeting = selectedMeeting {
-                    MeetingDetailContentView(meeting: selectedMeeting, onDelete: {
+                    MeetingDetailContentView(
+                        meeting: selectedMeeting,
+                        mergeCandidate: viewModel.previousMeeting(for: selectedMeeting),
+                        onMerge: { meeting in
+                            guard let merged = viewModel.mergeIntoPrevious(meeting) else { return false }
+                            self.selectedMeeting = merged
+                            return true
+                        },
+                        onDelete: {
                             // When a meeting is deleted from the detail view, clear the selection
                             self.selectedMeeting = nil
-                    })
+                        }
+                    )
                     .id(selectedMeeting.id) // Force recreation when selection changes
                 } else {
                     ContentUnavailableView(
@@ -229,12 +238,22 @@ struct MeetingDetailContentView: View {
     @StateObject private var viewModel: MeetingViewModel
     @StateObject private var recordingSessionManager = RecordingSessionManager.shared
     @State private var showDeleteAlert = false
+    @State private var showMergeAlert = false
     @State private var isEditing = false
     @State private var showCopyConfirmation = false
+    let mergeCandidate: Meeting?
+    let onMerge: (Meeting) -> Bool
     let onDelete: () -> Void
     
-    init(meeting: Meeting, onDelete: @escaping () -> Void) {
+    init(
+        meeting: Meeting,
+        mergeCandidate: Meeting?,
+        onMerge: @escaping (Meeting) -> Bool,
+        onDelete: @escaping () -> Void
+    ) {
         self._viewModel = StateObject(wrappedValue: MeetingViewModel(meeting: meeting))
+        self.mergeCandidate = mergeCandidate
+        self.onMerge = onMerge
         self.onDelete = onDelete
     }
     
@@ -276,6 +295,17 @@ struct MeetingDetailContentView: View {
                             Divider()
                         }
 
+                        if mergeCandidate != nil {
+                            Button {
+                                showMergeAlert = true
+                            } label: {
+                                Label("Merge into Previous Meeting", systemImage: "arrow.triangle.merge")
+                            }
+                            .disabled(recordingSessionManager.isRecording || recordingSessionManager.isProcessing)
+
+                            Divider()
+                        }
+
                         Button("Delete Meeting", role: .destructive) {
                             showDeleteAlert = true
                         }
@@ -461,6 +491,23 @@ struct MeetingDetailContentView: View {
         } message: {
             Text("Are you sure you want to delete this meeting? This action cannot be undone.")
         }
+        .alert("Merge into Previous Meeting?", isPresented: $showMergeAlert) {
+            Button("Merge", role: .destructive) {
+                // Prevent the disappearing detail view from auto-saving the
+                // continuation after the storage layer removes it.
+                viewModel.isDeleted = true
+                if !onMerge(viewModel.meeting) {
+                    viewModel.isDeleted = false
+                    viewModel.errorMessage = "The meetings could not be merged. Their original records and audio were kept."
+                }
+            }
+            Button("Cancel", role: .cancel) { }
+        } message: {
+            let previousTitle = mergeCandidate?.title.isEmpty == false
+                ? mergeCandidate?.title ?? "the previous meeting"
+                : "the previous meeting"
+            Text("This combines this meeting's transcript, notes, and saved audio into \"\(previousTitle)\", keeps the earlier meeting's title and time, then removes this continuation.")
+        }
         .onDisappear {
             // A failed recording may still be empty. Keep it until the user
             // explicitly deletes it so app updates cannot erase history.