Merge pull request #2 from argmaxinc/chen/OSSUpdate

Remove tracking, add background recording and stream mode config

Author: chen-argmax

Playground.xcodeproj/project.pbxproj

@@ -29,6 +29,7 @@
 		74F3B7C12E1CF4F400C544D1 /* AudioProcess.swift in Sources */ = {isa = PBXBuildFile; fileRef = 74F3B7C02E1CF4F400C544D1 /* AudioProcess.swift */; };
 		74F860942E29A9D20007163C /* ProcessTapper.swift in Sources */ = {isa = PBXBuildFile; fileRef = 74F860932E29A9D20007163C /* ProcessTapper.swift */; };
 		74F860962E2B19060007163C /* CoreAudioUtils.swift in Sources */ = {isa = PBXBuildFile; fileRef = 74F860952E2B19060007163C /* CoreAudioUtils.swift */; };
+		74F897792E4F9B130045252E /* TranscriptionModeSelection.swift in Sources */ = {isa = PBXBuildFile; fileRef = 74F897782E4F9B130045252E /* TranscriptionModeSelection.swift */; };
 /* End PBXBuildFile section */
 
 /* Begin PBXCopyFilesBuildPhase section */
@@ -79,6 +80,7 @@
 		74F3B7C02E1CF4F400C544D1 /* AudioProcess.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = AudioProcess.swift; sourceTree = "<group>"; };
 		74F860932E29A9D20007163C /* ProcessTapper.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ProcessTapper.swift; sourceTree = "<group>"; };
 		74F860952E2B19060007163C /* CoreAudioUtils.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = CoreAudioUtils.swift; sourceTree = "<group>"; };
+		74F897782E4F9B130045252E /* TranscriptionModeSelection.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = TranscriptionModeSelection.swift; sourceTree = "<group>"; };
 /* End PBXFileReference section */
 
 /* Begin PBXFrameworksBuildPhase section */
@@ -130,6 +132,7 @@
 		1677AFE42B5769E5008C61C0 /* Views */ = {
 			isa = PBXGroup;
 			children = (
+				74F897782E4F9B130045252E /* TranscriptionModeSelection.swift */,
 				74312CDD2E1DA46C000D994A /* StreamResultView.swift */,
 				1677AFE52B57704E008C61C0 /* ContentView.swift */,
 				74F3B7BB2E1C7C8B00C544D1 /* ToastMessage.swift */,
@@ -292,6 +295,7 @@
 				746E4C062E39874F009623D7 /* DefaultEnvInitializer.swift in Sources */,
 				1677AFC22B57618A008C61C0 /* Playground.swift in Sources */,
 				748BA5502E1B2EC6008DA1B8 /* StreamViewModel.swift in Sources */,
+				74F897792E4F9B130045252E /* TranscriptionModeSelection.swift in Sources */,
 				746E4C0A2E398757009623D7 /* PlaygroundEnvInitializer.swift in Sources */,
 				74F3B7BC2E1C7C8B00C544D1 /* ToastMessage.swift in Sources */,
 				74312CDE2E1DA46C000D994A /* StreamResultView.swift in Sources */,

Playground/Info.plist

@@ -9,5 +9,9 @@
 		<key>NSPrivacyAccessedAPIType</key>
 		<string>NSPrivacyAccessedAPICategoryUserDefaults</string>
 	</dict>
+	<key>UIBackgroundModes</key>
+	<array>
+		<string>audio</string>
+	</array>
 </dict>
 </plist>

Playground/ViewModels/StreamViewModel.swift

@@ -50,6 +50,9 @@ class StreamViewModel: ObservableObject {
     let sdkCoordinator: ArgmaxSDKCoordinator
 
     private var streamTasks: [Task<Void, Never>] = []
+    // Throttle guards to avoid overwhelming the UI with high-frequency updates
+    private var lastEnergyUpdateAt: TimeInterval = 0
+    private var lastHypothesisUpdateAtBySource: [String: TimeInterval] = [:]
 
     // Currently active streaming sources, set only in startTranscribing
     private var curActiveStreamSrcs: [any StreamSourceProtocol] = []
@@ -282,9 +285,16 @@ class StreamViewModel: ObservableObject {
     private func handleResult(_ result: LiveResult, for sourceId: String) {
         switch result {
         case .hypothesis(let text, _):
+            let now = Date().timeIntervalSince1970
+            let last = lastHypothesisUpdateAtBySource[sourceId] ?? 0
+            // Update at most 10 times per second per source
+            guard now - last >= 0.1 else { return }
+            lastHypothesisUpdateAtBySource[sourceId] = now
+            let trimmed = text.trimmingCharacters(in: .whitespacesAndNewlines)
+            guard trimmed != (isDeviceSource(sourceId) ? deviceResult?.hypothesisText : systemResult?.hypothesisText) else { return }
             updateStreamResult(sourceId: sourceId) { oldResult in
                 var newResult = oldResult
-                newResult.hypothesisText = text.trimmingCharacters(in: .whitespacesAndNewlines)
+                newResult.hypothesisText = trimmed
                 return newResult
             }
 
@@ -311,10 +321,23 @@ class StreamViewModel: ObservableObject {
     @MainActor
     private func updateAudioMetrics(for source: ArgmaxSource, audioData: [Float]) {
         if case .device = source.streamType, let whisperKitPro = self.sdkCoordinator.whisperKit {
+            let now = Date().timeIntervalSince1970
+            guard now - lastEnergyUpdateAt >= 0.1 else { return }
+            lastEnergyUpdateAt = now
+
+            // Limit the amount of energy samples passed to the UI for performance
+            let energies = whisperKitPro.audioProcessor.relativeEnergy
+            #if os(iOS)
+            let newBufferEnergy = Array(energies.suffix(256))
+            #else
+            let newBufferEnergy = energies
+            #endif
+            let sampleCount = whisperKitPro.audioProcessor.audioSamples.count
+
             updateStreamResult(sourceId: source.id) { oldResult in
                 var newResult = oldResult
-                newResult.bufferEnergy = whisperKitPro.audioProcessor.relativeEnergy
-                newResult.bufferSeconds = Double(whisperKitPro.audioProcessor.audioSamples.count) / Double(WhisperKit.sampleRate)
+                newResult.bufferEnergy = newBufferEnergy
+                newResult.bufferSeconds = Double(sampleCount) / Double(WhisperKit.sampleRate)
                 return newResult
             }
         }

Playground/Views/ContentView.swift

@@ -27,7 +27,7 @@ import Hub
 ///
 /// The view integrates with several key components:
 /// - `StreamViewModel`: Manages real-time audio streaming and transcription
-/// - `TranscribeViewModel`: Handles file-based transcription and recording workflows  
+/// - `TranscribeViewModel`: Handles file-based transcription and recording workflows
 /// - `ArgmaxSDKCoordinator`: Coordinates access to WhisperKit and SpeakerKit instances
 /// - Audio discovery services for device and process selection (macOS)
 ///
@@ -75,6 +75,8 @@ struct ContentView: View {
     @AppStorage("silenceThreshold") private var silenceThreshold: Double = 0.2
     @AppStorage("maxSilenceBufferLength") private var maxSilenceBufferLength: Double = 10.0
     @AppStorage("transcribeInterval") private var transcribeInterval: Double = 0.1
+    @AppStorage("minProcessInterval") private var minProcessInterval: Double = 0.0
+    @AppStorage("transcriptionMode") private var transcriptionModeRawValue: String = TranscriptionModeSelection.voiceTriggered.rawValue
     @AppStorage("useVAD") private var useVAD: Bool = true
     @AppStorage("tokenConfirmationsNeeded") private var tokenConfirmationsNeeded: Double = 2
     @AppStorage("concurrentWorkerCount") private var concurrentWorkerCount: Double = 4
@@ -91,6 +93,16 @@ struct ContentView: View {
     @AppStorage("fastLoadDecoderComputeUnits") private var fastLoadDecoderComputeUnits: MLComputeUnits = .cpuAndNeuralEngine
     #endif
     @AppStorage("trackingPermissionStatePro") private var trackingPermissionStateRawValue: Int = TrackingPermissionState.undetermined.rawValue
+    
+    /// Computed property to work with transcription mode as an enum
+    private var transcriptionMode: TranscriptionModeSelection {
+        get {
+            TranscriptionModeSelection(rawValue: transcriptionModeRawValue) ?? .voiceTriggered
+        }
+        set {
+            transcriptionModeRawValue = newValue.rawValue
+        }
+    }
 
     // MARK: Standard properties
 
@@ -139,7 +151,6 @@ struct ContentView: View {
 
     // MARK: Alerts
 
-    @State private var showReportingAlert = false
     @State private var showShortAudioWarningAlert: Bool = false
     @State private var showPermissionAlert: Bool = false
     @State private var permissionAlertMessage: String = ""
@@ -184,18 +195,6 @@ struct ContentView: View {
             set: { newValue in
                 trackingPermissionStateRawValue = newValue ? TrackingPermissionState.granted.rawValue : TrackingPermissionState.denied.rawValue
                 Logging.debug(newValue)
-
-                if newValue {
-                    sdkCoordinator.setupArgmax()
-                    analyticsLogger.configureIfNeeded()
-                } else {
-                    Task {
-                        if await ArgmaxSDK.enabled() {
-                            await ArgmaxSDK.close()
-                        }
-                        Logging.debug("Shutting down ArgmaxSDK")
-                    }
-                }
             }
         )
     }
@@ -348,18 +347,6 @@ struct ContentView: View {
                     #endif
                     .navigationSplitViewColumnWidth(min: 300, ideal: 350)
                     .padding(.horizontal)
-                    .alert(isPresented: $showReportingAlert) {
-                        Alert(
-                            title: Text("Performance Reporting"),
-                            message: Text("Help us catch bugs early and improve reliability by enabling reporting and performance monitoring. Required to enable experimental features. Learn more at [argmaxinc.com/privacy](https://www.argmaxinc.com/privacy)"),
-                            primaryButton: .default(Text("Enable reporting")) {
-                                updateTracking(state: .granted)
-                            },
-                            secondaryButton: .cancel(Text("Opt Out")) {
-                                updateTracking(state: .denied)
-                            }
-                        )
-                    }
                 } detail: {
                     VStack {
                         #if os(iOS)
@@ -448,12 +435,6 @@ struct ContentView: View {
             showWhisperKitComputeUnits = true
             speakerKitComputeUnitsExpanded = false
 
-            showReportingAlert = (trackingPermissionStateRawValue == 0) // undetermined
-            if trackingPermissionStateRawValue == TrackingPermissionState.granted.rawValue {
-                sdkCoordinator.setupArgmax()
-                analyticsLogger.configureIfNeeded()
-            }
-
             // Check if Pro models are supported on this OS version
             if #unavailable(macOS 15, iOS 18, watchOS 11, visionOS 2) {
                 showOSVersionAlert = true
@@ -1425,27 +1406,59 @@ struct ContentView: View {
             }
             .padding(.horizontal)
 
-            VStack {
-                Text("Silence Threshold")
+            Section(header: Text("Stream Mode Settings")) {
                 HStack {
-                    Slider(value: $silenceThreshold, in: 0...1, step: 0.05)
-                    Text(silenceThreshold.formatted(.number))
-                        .frame(width: 30)
-                    InfoButton("Relative silence threshold for the audio. \n Baseline is set by the quietest 100ms in the previous 2 seconds.")
+                    Picker("Mode", selection: Binding(
+                        get: { TranscriptionModeSelection(rawValue: transcriptionModeRawValue) ?? .voiceTriggered },
+                        set: { transcriptionModeRawValue = $0.rawValue }
+                    )) {
+                        ForEach(TranscriptionModeSelection.allCases) { mode in
+                            Text(mode.displayName).tag(mode)
+                        }
+                    }
+                    .pickerStyle(MenuPickerStyle())
+                    Spacer()
+                    InfoButton(transcriptionMode.description)
                 }
-            }
-            .padding(.horizontal)
-            
-            VStack {
-                Text("Max Silence Buffer Size")
-                HStack {
-                    Slider(value: $maxSilenceBufferLength, in: 10...60, step: 1)
-                    Text(maxSilenceBufferLength.formatted(.number))
-                        .frame(width: 30)
-                    InfoButton("Seconds of silence to buffer before audio is sent for transcription.")
+                
+                if transcriptionMode == .voiceTriggered {
+                    VStack {
+                        Text("Silence Threshold")
+                        HStack {
+                            Slider(value: $silenceThreshold, in: 0...1, step: 0.05)
+                            Text(silenceThreshold.formatted(.number.precision(.fractionLength(1))))
+                                .frame(width: 30)
+                                .lineLimit(1)
+                            InfoButton("Relative silence threshold for the audio. \n Baseline is set by the quietest 100ms in the previous 2 seconds.")
+                        }
+                    }
+                    .padding(.horizontal)
+                    
+                    VStack {
+                        Text("Max Silence Buffer Size")
+                        HStack {
+                            Slider(value: $maxSilenceBufferLength, in: 10...60, step: 1)
+                            Text(maxSilenceBufferLength.formatted(.number.precision(.fractionLength(0))))
+                                .frame(width: 30)
+                                .lineLimit(1)
+                            InfoButton("Seconds of silence to buffer before audio is sent for transcription.")
+                        }
+                    }
+                    .padding(.horizontal)
+                    
+                    VStack {
+                        Text("Min Process Interval")
+                        HStack {
+                            Slider(value: $minProcessInterval, in: 0...15, step: 1)
+                            Text(minProcessInterval.formatted(.number.precision(.fractionLength(0))))
+                                .frame(width: 30)
+                                .lineLimit(1)
+                            InfoButton("Minimum interval the incoming stream data is fed to transcription pipeline.")
+                        }
+                    }
+                    .padding(.horizontal)
                 }
             }
-            .padding(.horizontal)
 
             VStack {
                 Text("Transcribe Interval")
@@ -1458,21 +1471,6 @@ struct ContentView: View {
             }
             .padding(.horizontal)
 
-            Section(header: Text("Performance Reporting")) {
-                VStack(alignment: .leading) {
-                    HStack {
-                        Text("Enable Reporting")
-                        InfoButton("Help us catch bugs early and improve reliability by enabling reporting and performance monitoring.")
-                        Spacer()
-                        Toggle("", isOn: trackingPermissionBinding)
-                    }
-                    Link(destination: URL(string: "https://www.argmaxinc.com/privacy")!) {
-                        Text("Learn more at argmaxinc.com/privacy")
-                    }
-                }
-                .padding(.horizontal)
-                .padding(.top)
-            }
             Section(header: Text("Diarization Settings")) {
                 HStack {
                     Picker("Diarization", selection: $diarizationMode) {
@@ -2074,11 +2072,21 @@ struct ContentView: View {
                     isRecording = true
                 }
 
+                let streamMode: StreamTranscriptionMode
+                switch transcriptionMode {
+                case .alwaysOn:
+                    streamMode = .alwaysOn
+                case .voiceTriggered:
+                    streamMode = .voiceTriggered(silenceThreshold: Float(silenceThreshold), maxBufferLength: Float(maxSilenceBufferLength), minProcessInterval: Float(minProcessInterval))
+                case .batteryOptimized:
+                    streamMode = .batteryOptimized
+                }
+                
                 try await streamViewModel.startTranscribing(
                     options: DecodingOptionsPro(
                         base: decodingOptions,
                         transcribeInterval: transcribeInterval,
-                        streamTranscriptionMode: .voiceTriggered(silenceThreshold: Float(silenceThreshold), maxBufferLength: Float(maxSilenceBufferLength))
+                        streamTranscriptionMode: streamMode
                     )
                 )
             } catch {
@@ -2188,6 +2196,7 @@ struct ContentView: View {
             "compression_check_window": "\(compressionCheckWindow)",
             "sample_length": "\(sampleLength)",
             "silence_threshold": "\(silenceThreshold)",
+            "transcription_mode": "\(transcriptionMode.rawValue)",
             "use_vad": "\(useVAD)",
             "token_confirmations_needed": "\(tokenConfirmationsNeeded)",
             "chunking_strategy": "\(chunkingStrategy)",

Playground/Views/StreamResultView.swift

@@ -53,14 +53,13 @@ struct StreamResultLine: View {
                     .id("bottom")
                 }
                 .onChange(of: result.confirmedText) {
-                    withAnimation(.easeOut(duration: 0.3)) {
+                    withAnimation(.easeOut(duration: 0.15)) {
                         proxy.scrollTo("bottom", anchor: .bottom)
                     }
                 }
+                // Avoid animating on every hypothesis token; keep scroll position but don't animate
                 .onChange(of: result.hypothesisText) {
-                    withAnimation(.easeOut(duration: 0.3)) {
-                        proxy.scrollTo("bottom", anchor: .bottom)
-                    }
+                    proxy.scrollTo("bottom", anchor: .bottom)
                 }
             }
         }

Playground/Views/TranscriptionModeSelection.swift

@@ -0,0 +1,30 @@
+/// Enumeration representing the available transcription modes for stream processing.
+enum TranscriptionModeSelection: String, CaseIterable, Identifiable {
+    case alwaysOn = "alwaysOn"
+    case voiceTriggered = "voiceTriggered"
+    case batteryOptimized = "batteryOptimized"
+    
+    var id: String { rawValue }
+    
+    var displayName: String {
+        switch self {
+        case .alwaysOn:
+            return "Always-On"
+        case .voiceTriggered:
+            return "Voice-Triggered"
+        case .batteryOptimized:
+            return "Battery-Optimized"
+        }
+    }
+    
+    var description: String {
+        switch self {
+        case .alwaysOn:
+            return "Continuous real-time transcription with lowest latency. Uses more system resources."
+        case .voiceTriggered:
+            return "Processes only audio above energy threshold. Conserves battery while staying responsive."
+        case .batteryOptimized:
+            return "Intelligent streaming with dynamic optimizations for maximum battery life."
+        }
+    }
+}

Playground/Views/VoiceEnergyView.swift

@@ -1,7 +1,7 @@
 import Foundation
 import SwiftUI
 
-/// A SwiftUI view that visualizes audio buffer energy levels with threshold-based color coding.  
+/// A SwiftUI view that visualizes audio buffer energy levels with threshold-based color coding.
 /// This component provides real-time visual feedback for audio input levels and voice activity detection.
 ///
 /// ## Features
@@ -28,14 +28,12 @@ struct VoiceEnergyView: View {
 
     var body: some View {
         ScrollView(.horizontal) {
-            HStack(spacing: 1) {
-                ForEach(Array(bufferEnergy.enumerated())[0...], id: \.element) { _, energy in
-                    ZStack {
-                        RoundedRectangle(cornerRadius: 2)
-                            .frame(width: 2, height: CGFloat(energy) * 24)
-                    }
-                    .frame(maxHeight: 24)
-                    .background(energy > Float(silenceThreshold) ? Color.green.opacity(0.2) : Color.red.opacity(0.2))
+            LazyHStack(spacing: 1) {
+                ForEach(Array(bufferEnergy.enumerated()), id: \.offset) { _, energy in
+                    RoundedRectangle(cornerRadius: 2)
+                        .frame(width: 2, height: max(0, min(CGFloat(energy), 1)) * 24)
+                        .frame(maxHeight: 24)
+                        .background(energy > Float(silenceThreshold) ? Color.green.opacity(0.2) : Color.red.opacity(0.2))
                 }
             }
         }