fix: allow recording while meetings process

Meetingnotes.xcodeproj/project.pbxproj

@@ -276,7 +276,7 @@
 				CODE_SIGN_IDENTITY = "Apple Development";
 				CODE_SIGN_STYLE = Automatic;
 				COMBINE_HIDPI_IMAGES = YES;
-				CURRENT_PROJECT_VERSION = 42;
+				CURRENT_PROJECT_VERSION = 44;
 				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 = 1.1.30;
+				MARKETING_VERSION = 1.1.32;
 				ONLY_ACTIVE_ARCH = NO;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;
@@ -312,7 +312,7 @@
 				CODE_SIGN_IDENTITY = "Apple Development";
 				CODE_SIGN_STYLE = Automatic;
 				COMBINE_HIDPI_IMAGES = YES;
-				CURRENT_PROJECT_VERSION = 42;
+				CURRENT_PROJECT_VERSION = 44;
 				DEVELOPMENT_ASSET_PATHS = "\"meetingnotes/Preview Content\"";
 				DEVELOPMENT_TEAM = G9LVHZAJNX;
 				ENABLE_HARDENED_RUNTIME = YES;
@@ -326,7 +326,7 @@
 					"@executable_path/../Frameworks",
 				);
 				MACOSX_DEPLOYMENT_TARGET = 15.0;
-				MARKETING_VERSION = 1.1.30;
+				MARKETING_VERSION = 1.1.32;
 				ONLY_ACTIVE_ARCH = YES;
 				OTHER_SWIFT_FLAGS = "$(inherited) -D ENABLE_TCC_SPI";
 				PRODUCT_BUNDLE_IDENTIFIER = net.jamesbone.meetingnotes;

meetingnotes/Managers/AudioManager.swift

@@ -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
@@ -72,7 +73,7 @@ final class AudioManager: NSObject, ObservableObject {
     private var recordingStartedAt = Date()
     private var systemDiagnostics = SystemCaptureDiagnostics()
 
-    private override init() {
+    override init() {
         super.init()
         observeAudioEngine()
     }
@@ -443,18 +444,70 @@ 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,
+                      let previousNow = pendingNow 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,
+                    previousHostTimestamp: previousNow,
+                    currentHostTimestamp: inNow.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 +528,7 @@ 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 +545,67 @@ 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,
+        previousHostTimestamp: AudioTimeStamp,
+        currentHostTimestamp: AudioTimeStamp
+    ) -> Double {
+        let sampleTimeDelta = currentTimestamp.mSampleTime - previousTimestamp.mSampleTime
+        guard advertisedSampleRate > 0, previousFrameLength > 0 else {
+            systemDiagnostics.inputRateInference = "advertised (invalid advertised rate or frame count)"
+            return advertisedSampleRate
+        }
+
+        guard currentHostTimestamp.mHostTime > previousHostTimestamp.mHostTime else {
+            systemDiagnostics.inputRateInference = "advertised (host timestamps unavailable)"
+            return advertisedSampleRate
+        }
+        let hostTimeDelta = currentHostTimestamp.mHostTime - previousHostTimestamp.mHostTime
+        let hostNanoseconds = AudioConvertHostTimeToNanos(hostTimeDelta)
+        guard hostNanoseconds > 0 else {
+            systemDiagnostics.inputRateInference = "advertised (host-time conversion failed)"
+            return advertisedSampleRate
+        }
+
+        let inferredSampleRate = Double(previousFrameLength) * 1_000_000_000 / Double(hostNanoseconds)
+        let plausibleRange = (advertisedSampleRate * 0.25)...(advertisedSampleRate * 1.25)
+        guard inferredSampleRate.isFinite, plausibleRange.contains(inferredSampleRate) else {
+            systemDiagnostics.inputRateInference = String(
+                format: "advertised (invalid host inference %.3f from %u frames / %llu ns; sampleTimeDelta=%.3f)",
+                inferredSampleRate,
+                previousFrameLength,
+                hostNanoseconds,
+                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: "source=hostTime,advertised=%.3f,inferred=%.3f,selected=%.3f,previousFrames=%u,hostNanoseconds=%llu,sampleTimeDelta=%.3f",
+            advertisedSampleRate,
+            inferredSampleRate,
+            selectedSampleRate,
+            previousFrameLength,
+            hostNanoseconds,
+            sampleTimeDelta
+        )
+        return selectedSampleRate
+    }
+
     private func copyAudioBuffer(
         from inputData: UnsafePointer<AudioBufferList>,
         format: AVAudioFormat
@@ -863,6 +971,7 @@ 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)",

meetingnotes/Managers/RecordingSessionManager.swift

@@ -2,6 +2,12 @@ import Foundation
 import SwiftUI
 import Combine
 
+struct RecordingCompletion {
+    let chunks: [TranscriptChunk]
+    let recoveryAudioFolderName: String?
+    let transcriptionError: String?
+}
+
 /// Manages recording sessions at the app level to persist across navigation
 @MainActor
 class RecordingSessionManager: ObservableObject {
@@ -14,9 +20,11 @@ class RecordingSessionManager: ObservableObject {
     @Published var errorMessage: String?
     @Published var activeRecordingTranscriptChunksUpdated: [TranscriptChunk] = []
     
-    private let audioManager = AudioManager.shared
+    private var audioManager = AudioManager.shared
     private var cancellables = Set<AnyCancellable>()
+    private var audioManagerCancellables = Set<AnyCancellable>()
     private let transcriptUpdateSubject = PassthroughSubject<[TranscriptChunk], Never>()
+    private var processingMeetingIds = Set<UUID>()
     
     // Store transcript chunks for the active recording session
     private var activeRecordingTranscriptChunks: [TranscriptChunk] = []
@@ -27,24 +35,19 @@ class RecordingSessionManager: ObservableObject {
     }
     
     private func setupAudioManagerBindings() {
+        audioManagerCancellables.removeAll()
         // Bind to audio manager state
         audioManager.$isRecording
             .sink { [weak self] isRecording in
                 self?.isRecording = isRecording
             }
-            .store(in: &cancellables)
-        
-        audioManager.$isProcessing
-            .sink { [weak self] isProcessing in
-                self?.isProcessing = isProcessing
-            }
-            .store(in: &cancellables)
+            .store(in: &audioManagerCancellables)
 
         audioManager.$errorMessage
             .sink { [weak self] errorMessage in
                 self?.errorMessage = errorMessage
             }
-            .store(in: &cancellables)
+            .store(in: &audioManagerCancellables)
         
         // When transcript chunks change, store them for the active recording and send to debouncer
         audioManager.$transcriptChunks
@@ -55,7 +58,7 @@ class RecordingSessionManager: ObservableObject {
                 
                 self.transcriptUpdateSubject.send(newChunks)
             }
-            .store(in: &cancellables)
+            .store(in: &audioManagerCancellables)
     }
     
     private func setupDebouncedSaving() {
@@ -84,26 +87,40 @@ class RecordingSessionManager: ObservableObject {
         audioManager.startRecording(for: meetingId)
     }
     
-    func stopRecording() async -> [TranscriptChunk] {
+    func stopRecording() async -> RecordingCompletion {
         print("🛑 Stopping recording for meeting: \(activeMeetingId?.uuidString ?? "unknown")")
         
         guard let meetingId = activeMeetingId else {
             audioManager.cancelRecording()
             recordingStartedAt = nil
-            return []
+            return RecordingCompletion(chunks: [], recoveryAudioFolderName: nil, transcriptionError: nil)
         }
+
+        let completedAudioManager = audioManager
+        audioManager = AudioManager()
+        setupAudioManagerBindings()
+        activeMeetingId = nil
         recordingStartedAt = nil
-        let chunks = await audioManager.stopRecordingAndTranscribe()
-        activeRecordingTranscriptChunks = chunks
-        activeRecordingTranscriptChunksUpdated = chunks
+        processingMeetingIds.insert(meetingId)
+        isProcessing = true
+        defer {
+            processingMeetingIds.remove(meetingId)
+            isProcessing = !processingMeetingIds.isEmpty
+        }
+
+        let chunks = await completedAudioManager.stopRecordingAndTranscribe()
+        let completion = RecordingCompletion(
+            chunks: chunks,
+            recoveryAudioFolderName: completedAudioManager.lastRecoveryAudioFolderName,
+            transcriptionError: completedAudioManager.errorMessage
+        )
         updateActiveMeetingTranscript(
             meetingId: meetingId,
             chunks: chunks,
-            transcriptionError: audioManager.errorMessage
+            recoveryAudioFolderName: completion.recoveryAudioFolderName,
+            transcriptionError: completion.transcriptionError
         )
-        activeMeetingId = nil
-        activeRecordingTranscriptChunks = []
-        return chunks
+        return completion
     }
         
     func cancelRecording() {
@@ -117,13 +134,14 @@ class RecordingSessionManager: ObservableObject {
         return isRecording && activeMeetingId == meetingId
     }
 
-    var lastRecoveryAudioFolderName: String? {
-        audioManager.lastRecoveryAudioFolderName
+    func isProcessingMeeting(_ meetingId: UUID) -> Bool {
+        processingMeetingIds.contains(meetingId)
     }
-    
+
     private func updateActiveMeetingTranscript(
         meetingId: UUID,
         chunks: [TranscriptChunk],
+        recoveryAudioFolderName: String? = nil,
         transcriptionError: String? = nil
     ) {
         // Load all meetings
@@ -132,7 +150,7 @@ class RecordingSessionManager: ObservableObject {
         // Find and update the active meeting
         if let index = meetings.firstIndex(where: { $0.id == meetingId }) {
             meetings[index].transcriptChunks = chunks
-            if let recoveryAudioFolderName = lastRecoveryAudioFolderName {
+            if let recoveryAudioFolderName {
                 meetings[index].recoveryAudioFolderName = recoveryAudioFolderName
             }
             meetings[index].transcriptionError = transcriptionError

meetingnotes/Services/CoderAPIClient.swift

@@ -30,6 +30,7 @@ enum CoderAPIError: LocalizedError {
     case missingModel(String)
     case invalidResponse
     case serviceError(Int, String)
+    case audioPreparationFailed(String, String)
 
     var errorDescription: String? {
         switch self {
@@ -43,6 +44,8 @@ enum CoderAPIError: LocalizedError {
             return "Coder returned an invalid response."
         case .serviceError(let status, let message):
             return "Coder request failed (\(status)): \(message)"
+        case .audioPreparationFailed(let filename, let message):
+            return "Could not prepare \(filename) for transcription: \(message)"
         }
     }
 }
@@ -179,7 +182,12 @@ final class CoderAPIClient {
         let selectedModel = model.trimmingCharacters(in: .whitespacesAndNewlines)
         guard !selectedModel.isEmpty else { throw CoderAPIError.missingModel("transcription") }
         let apiKey = try requiredAPIKey(KeychainHelper.shared.getCoderAPIKey() ?? "")
-        let chunks = try makeAudioChunks(from: fileURL, preserveSpeakerIdentity: diarization)
+        let chunks: [AudioChunk]
+        do {
+            chunks = try makeAudioChunks(from: fileURL, preserveSpeakerIdentity: diarization)
+        } catch {
+            throw CoderAPIError.audioPreparationFailed(fileURL.lastPathComponent, error.localizedDescription)
+        }
         defer {
             for chunk in chunks where chunk.isTemporary {
                 try? FileManager.default.removeItem(at: chunk.url)
@@ -260,13 +268,38 @@ final class CoderAPIClient {
            let size = attributes[.size] as? NSNumber {
             request.setValue(size.stringValue, forHTTPHeaderField: "Content-Length")
         }
-        let (data, response) = try await transcriptionSession.upload(for: request, fromFile: bodyURL)
+        let (data, response) = try await uploadTranscription(request: request, bodyURL: bodyURL)
         try validate(response: response, data: data)
         let decoded = try JSONDecoder().decode(TranscriptionResponse.self, from: data)
         let segments = decoded.segments ?? segments(from: decoded.words ?? [])
         return Transcription(text: decoded.text, segments: segments)
     }
 
+    private func uploadTranscription(request: URLRequest, bodyURL: URL) async throws -> (Data, URLResponse) {
+        let retryDelays: [UInt64] = [3, 10]
+        for attempt in 0...retryDelays.count {
+            do {
+                return try await transcriptionSession.upload(for: request, fromFile: bodyURL)
+            } catch {
+                guard attempt < retryDelays.count, isRetryableTranscriptionError(error) else {
+                    throw error
+                }
+                try await Task.sleep(nanoseconds: retryDelays[attempt] * 1_000_000_000)
+            }
+        }
+        throw CoderAPIError.invalidResponse
+    }
+
+    private func isRetryableTranscriptionError(_ error: Error) -> Bool {
+        guard let urlError = error as? URLError else { return false }
+        switch urlError.code {
+        case .networkConnectionLost, .cannotConnectToHost, .timedOut:
+            return true
+        default:
+            return false
+        }
+    }
+
     private func makeAudioChunks(from fileURL: URL, preserveSpeakerIdentity: Bool) throws -> [AudioChunk] {
         let input = try AVAudioFile(forReading: fileURL)
         let format = input.processingFormat

meetingnotes/Services/LocalAPIServer.swift

@@ -182,13 +182,13 @@ private enum LocalAPIRouter {
 }
 
 private enum LocalRecordingError: LocalizedError {
-    case processing
+    case stopping
     case saveFailed
 
     var errorDescription: String? {
         switch self {
-        case .processing:
-            return "The previous meeting is still processing."
+        case .stopping:
+            return "The current recording is still stopping."
         case .saveFailed:
             return "Could not create a meeting for this recording."
         }
@@ -208,12 +208,12 @@ private final class LocalRecordingController {
 
     func statusPayload() -> [String: Any] {
         let state: String
-        if isStopping || recordingManager.isProcessing {
-            state = "processing"
-        } else if recordingManager.isRecording {
+        if recordingManager.isRecording {
             state = "recording"
         } else if recordingManager.activeMeetingId != nil {
             state = "starting"
+        } else if isStopping || recordingManager.isProcessing {
+            state = "processing"
         } else {
             state = "idle"
         }
@@ -245,8 +245,8 @@ private final class LocalRecordingController {
             isStopping = false
             return statusPayload()
         }
-        if isStopping || recordingManager.isProcessing {
-            throw LocalRecordingError.processing
+        if isStopping {
+            throw LocalRecordingError.stopping
         }
         if recordingManager.activeMeetingId != nil {
             return statusPayload()
@@ -295,15 +295,16 @@ private final class LocalRecordingController {
     private func finishRecording() async {
         pendingStopTask = nil
         let meetingID = recordingManager.activeMeetingId
-        let chunks = await recordingManager.stopRecording()
+        isStopping = false
+        let completion = await recordingManager.stopRecording()
         guard let meetingID,
               var meeting = LocalStorageManager.shared.loadMeetings().first(where: { $0.id == meetingID }) else {
-            isStopping = false
             return
         }
 
-        meeting.transcriptChunks = chunks
-        meeting.recoveryAudioFolderName = recordingManager.lastRecoveryAudioFolderName
+        meeting.transcriptChunks = completion.chunks
+        meeting.recoveryAudioFolderName = completion.recoveryAudioFolderName
+        meeting.transcriptionError = completion.transcriptionError
         let templates = LocalStorageManager.shared.loadTemplates()
         if meeting.templateId == nil {
             meeting.templateId = LocalStorageManager.shared.preferredTemplateID(in: templates)
@@ -352,7 +353,6 @@ private final class LocalRecordingController {
                 NotificationCenter.default.post(name: .meetingSaved, object: meeting)
             }
         }
-        isStopping = false
     }
 }
 

meetingnotes/ViewModels/MeetingViewModel.swift

@@ -47,7 +47,7 @@ class MeetingViewModel: ObservableObject {
 
     var isProcessing: Bool {
         return isRetryingTranscription ||
-            (recordingSessionManager.isProcessing && recordingSessionManager.activeMeetingId == meeting.id)
+            recordingSessionManager.isProcessingMeeting(meeting.id)
     }
 
     var canRetryTranscription: Bool {
@@ -137,13 +137,11 @@ class MeetingViewModel: ObservableObject {
             self.meeting.transcriptChunks = recordingSessionManager.getTranscriptChunks(for: meeting.id)
         }
 
-        // Listen for final transcript updates for this meeting.
+        // Listen for transcript updates emitted while this meeting is recording.
         recordingSessionManager.$activeRecordingTranscriptChunksUpdated
             .dropFirst()
             .sink { [weak self] updatedChunks in
                 guard let self = self else { return }
-                // Stop publishes the final chunks after capture has ended but
-                // before the active meeting is cleared.
                 if recordingSessionManager.activeMeetingId == self.meeting.id {
                     self.meeting.transcriptChunks = updatedChunks
                 }
@@ -223,10 +221,11 @@ class MeetingViewModel: ObservableObject {
     func stopRecording() {
         isStartingRecording = true
         Task {
-            let chunks = await recordingSessionManager.stopRecording()
-            meeting.transcriptChunks = chunks
-            meeting.recoveryAudioFolderName = recordingSessionManager.lastRecoveryAudioFolderName
-            meeting.transcriptionError = recordingSessionManager.errorMessage
+            let completion = await recordingSessionManager.stopRecording()
+            meeting.transcriptChunks = completion.chunks
+            meeting.recoveryAudioFolderName = completion.recoveryAudioFolderName
+            meeting.transcriptionError = completion.transcriptionError
+            errorMessage = completion.transcriptionError
             refreshRecoveryAudioFolder()
             saveMeeting()
             if !meeting.formattedTranscript.isEmpty {