Skip to content

Fastest and only macOS Dictation app with on-device STT and custom trained AI enhancement model. Windows, iOS testflight! Android, Linux waitlist open! A local Wispr Flow alternative. Follow https://x.com/fluidvoiceapp for more

fix(dictionary): handle fallback and dismissals

altic-dev28616df9a455
diff --git a/Sources/Fluid/ContentView.swift b/Sources/Fluid/ContentView.swift
index 48c016f2..3b01983d 100644
--- a/Sources/Fluid/ContentView.swift
+++ b/Sources/Fluid/ContentView.swift
@@ -3528,7 +3528,7 @@ struct ContentView: View {
if DictionaryCorrectionOverlayController.shared.isPresented {
DebugLogger.shared.debug("Cancel shortcut: closing dictionary suggestion", source: "ContentView")
- DictionaryCorrectionOverlayController.shared.hide()
+ DictionaryCorrectionOverlayController.shared.dismiss()
return true
}
diff --git a/Sources/Fluid/Services/ASRService.swift b/Sources/Fluid/Services/ASRService.swift
index c780641b..a10cc8e4 100644
--- a/Sources/Fluid/Services/ASRService.swift
+++ b/Sources/Fluid/Services/ASRService.swift
@@ -101,6 +101,7 @@ final class ASRService: ObservableObject {
@Published private(set) var isCancellingModelDownload: Bool = false
@Published private(set) var isDictionaryTrainingCaptureActive: Bool = false
private(set) var lastDictionaryTrainingResult: ASRTranscriptionResult?
+ private(set) var dictionaryTrainingAudioGeneration = 0
private var isStarting: Bool = false // Guard against re-entrant start() calls
private var hasCompletedFirstTranscription: Bool = false // Track if model has warmed up with first transcription
@@ -896,6 +897,7 @@ final class ASRService: ObservableObject {
await self.stopStreamingTimerAndAwait()
guard self.isRunning else { return }
self.audioBuffer.clear(keepingCapacity: true)
+ self.dictionaryTrainingAudioGeneration &+= 1
self.lastProcessedSampleCount = 0
self.benchmarkLastChunkSampleCount = 0
(self.transcriptionProvider as? FluidAudioProvider)?.resetStreamingPreviewCache()
diff --git a/Sources/Fluid/Services/DictionaryTrainingEndpointMonitor.swift b/Sources/Fluid/Services/DictionaryTrainingEndpointMonitor.swift
index 39df201d..5cf172da 100644
--- a/Sources/Fluid/Services/DictionaryTrainingEndpointMonitor.swift
+++ b/Sources/Fluid/Services/DictionaryTrainingEndpointMonitor.swift
@@ -1,5 +1,24 @@
import Foundation
+struct DictionaryTrainingAudioCursor {
+ private(set) var sampleOffset = 0
+ private var generation: Int
+
+ init(generation: Int) {
+ self.generation = generation
+ }
+
+ mutating func synchronize(generation: Int) {
+ guard generation != self.generation else { return }
+ self.generation = generation
+ self.sampleOffset = 0
+ }
+
+ mutating func consume(_ sampleCount: Int) {
+ self.sampleOffset += sampleCount
+ }
+}
+
@MainActor
final class DictionaryTrainingEndpointMonitor {
static let shared = DictionaryTrainingEndpointMonitor()
@@ -42,19 +61,20 @@ final class DictionaryTrainingEndpointMonitor {
Task { await detector.endSession(detectorSession) }
}
- var sampleOffset = 0
+ var cursor = DictionaryTrainingAudioCursor(generation: asr.dictionaryTrainingAudioGeneration)
while !Task.isCancelled {
guard asr.isRunning, asr.isDictionaryTrainingCaptureActive else { return }
+ cursor.synchronize(generation: asr.dictionaryTrainingAudioGeneration)
let chunk = asr.dictionaryTrainingAudioChunk(
- at: sampleOffset,
+ at: cursor.sampleOffset,
count: DictionaryTrainingEndpointDetector.chunkSize
)
guard !chunk.isEmpty else {
try await Task.sleep(nanoseconds: 40_000_000)
continue
}
- sampleOffset += chunk.count
+ cursor.consume(chunk.count)
guard let event = try await detector.process(
chunk,
diff --git a/Sources/Fluid/Views/AutomaticDictionaryCorrectionOverlay.swift b/Sources/Fluid/Views/AutomaticDictionaryCorrectionOverlay.swift
index bc1f03ad..4b2d8403 100644
--- a/Sources/Fluid/Views/AutomaticDictionaryCorrectionOverlay.swift
+++ b/Sources/Fluid/Views/AutomaticDictionaryCorrectionOverlay.swift
@@ -53,8 +53,7 @@ final class DictionaryCorrectionOverlayController {
session: session,
displayDuration: Double(Self.displayDurationNanoseconds) / 1_000_000_000,
onDismiss: { [weak self] in
- self?.reportOutcome(.dismissed)
- self?.hide()
+ self?.dismiss()
}
)
@@ -121,6 +120,11 @@ final class DictionaryCorrectionOverlayController {
}
}
+ func dismiss() {
+ self.reportOutcome(.dismissed)
+ self.hide()
+ }
+
private func keepVisible() {
self.dismissTask?.cancel()
self.dismissTask = nil
diff --git a/Tests/FluidDictationIntegrationTests/DictationE2ETests.swift b/Tests/FluidDictationIntegrationTests/DictationE2ETests.swift
index daa244a3..f8f031b5 100644
--- a/Tests/FluidDictationIntegrationTests/DictationE2ETests.swift
+++ b/Tests/FluidDictationIntegrationTests/DictationE2ETests.swift
@@ -885,6 +885,16 @@ final class DictationE2ETests: XCTestCase {
XCTAssertFalse(SettingsStore.SpeechModel.cohereTranscribeSixBit.supportsPronunciationMatching)
}
+ func testDictionaryTrainingAudioCursorResetsAfterBufferGenerationChange() {
+ var cursor = DictionaryTrainingAudioCursor(generation: 4)
+ cursor.consume(1_600)
+ cursor.synchronize(generation: 4)
+ XCTAssertEqual(cursor.sampleOffset, 1_600)
+
+ cursor.synchronize(generation: 5)
+ XCTAssertEqual(cursor.sampleOffset, 0)
+ }
+
func testProgressiveDownloaderRetainsFileByMovingIt() throws {
let source = FileManager.default.temporaryDirectory
.appendingPathComponent("FluidVoiceDownloadSource-\(UUID().uuidString)")