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

feat(speech): Allow downloading models without changing active selection

altic-dev7a1d69cfed7d
diff --git a/Sources/Fluid/Persistence/SettingsStore.swift b/Sources/Fluid/Persistence/SettingsStore.swift
index bd963290..a3f6494c 100644
--- a/Sources/Fluid/Persistence/SettingsStore.swift
+++ b/Sources/Fluid/Persistence/SettingsStore.swift
@@ -1453,7 +1453,7 @@ final class SettingsStore: ObservableObject {
case whisperBase = "whisper-base"
case whisperSmall = "whisper-small"
case whisperMedium = "whisper-medium"
- case whisperLargeTurbo = "whisper-large-turbo"
+ // case whisperLargeTurbo = "whisper-large-turbo" // buggy - so removed temporarily
case whisperLarge = "whisper-large"
var id: String { rawValue }
@@ -1470,7 +1470,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return "Whisper Base"
case .whisperSmall: return "Whisper Small"
case .whisperMedium: return "Whisper Medium"
- case .whisperLargeTurbo: return "Whisper Large Turbo"
+ // case .whisperLargeTurbo: return "Whisper Large Turbo" // buggy - so removed temporarily
case .whisperLarge: return "Whisper Large"
}
}
@@ -1481,7 +1481,7 @@ final class SettingsStore: ObservableObject {
case .parakeetTDTv2: return "English Only (Higher Accuracy)"
case .appleSpeech: return "System Languages"
case .appleSpeechAnalyzer: return "EN, ES, FR, DE, IT, JA, KO, PT, ZH"
- case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, .whisperLargeTurbo, .whisperLarge:
+ case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, /* .whisperLargeTurbo, */ .whisperLarge: // buggy - so removed temporarily
return "99 Languages"
}
}
@@ -1496,7 +1496,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return "~142 MB"
case .whisperSmall: return "~466 MB"
case .whisperMedium: return "~1.5 GB"
- case .whisperLargeTurbo: return "~1.6 GB"
+ // case .whisperLargeTurbo: return "~1.6 GB" // buggy - so removed temporarily
case .whisperLarge: return "~2.9 GB"
}
}
@@ -1522,7 +1522,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return "ggml-base.bin"
case .whisperSmall: return "ggml-small.bin"
case .whisperMedium: return "ggml-medium.bin"
- case .whisperLargeTurbo: return "ggml-large-v3-turbo.bin"
+ // case .whisperLargeTurbo: return "ggml-large-v3-turbo.bin" // buggy - so removed temporarily
case .whisperLarge: return "ggml-large-v3.bin"
default: return nil
}
@@ -1535,7 +1535,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return "base"
case .whisperSmall: return "small"
case .whisperMedium: return "medium"
- case .whisperLargeTurbo: return "large-v3-turbo"
+ // case .whisperLargeTurbo: return "large-v3-turbo" // buggy - so removed temporarily
case .whisperLarge: return "large-v3"
default: return nil
}
@@ -1588,7 +1588,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return "Standard Choice"
case .whisperSmall: return "Balanced Speed & Accuracy"
case .whisperMedium: return "Medium Quality"
- case .whisperLargeTurbo: return "Higher Quality but Faster"
+ // case .whisperLargeTurbo: return "Higher Quality but Faster" // buggy - so removed temporarily
case .whisperLarge: return "Maximum Accuracy"
}
}
@@ -1612,8 +1612,8 @@ final class SettingsStore: ObservableObject {
return "Better accuracy than Base. Moderate resource usage."
case .whisperMedium:
return "High accuracy for demanding tasks. Requires more memory."
- case .whisperLargeTurbo:
- return "Near-maximum accuracy with optimized speed."
+ // case .whisperLargeTurbo: // buggy - so removed temporarily
+ // return "Near-maximum accuracy with optimized speed."
case .whisperLarge:
return "Best possible accuracy. Large download and memory usage."
}
@@ -1630,7 +1630,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return 4
case .whisperSmall: return 3
case .whisperMedium: return 2
- case .whisperLargeTurbo: return 3
+ // case .whisperLargeTurbo: return 3 // buggy - so removed temporarily
case .whisperLarge: return 1
}
}
@@ -1646,7 +1646,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return 3
case .whisperSmall: return 4
case .whisperMedium: return 4
- case .whisperLargeTurbo: return 5
+ // case .whisperLargeTurbo: return 5 // buggy - so removed temporarily
case .whisperLarge: return 5
}
}
@@ -1662,7 +1662,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return 0.80
case .whisperSmall: return 0.60
case .whisperMedium: return 0.40
- case .whisperLargeTurbo: return 0.65
+ // case .whisperLargeTurbo: return 0.65 // buggy - so removed temporarily
case .whisperLarge: return 0.20
}
}
@@ -1678,7 +1678,7 @@ final class SettingsStore: ObservableObject {
case .whisperBase: return 0.60
case .whisperSmall: return 0.70
case .whisperMedium: return 0.80
- case .whisperLargeTurbo: return 0.95
+ // case .whisperLargeTurbo: return 0.95 // buggy - so removed temporarily
case .whisperLarge: return 1.00
}
}
@@ -1707,7 +1707,7 @@ final class SettingsStore: ObservableObject {
/// Large Whisper models are too slow for streaming, so they only do final transcription on stop.
var supportsStreaming: Bool {
switch self {
- case .whisperMedium, .whisperLarge, .whisperLargeTurbo:
+ case .whisperMedium, .whisperLarge: // .whisperLargeTurbo - buggy - so removed temporarily
return false // Too slow for real-time chunk processing
default:
return true // All other models support streaming
@@ -1728,7 +1728,7 @@ final class SettingsStore: ObservableObject {
return .nvidia
case .appleSpeech, .appleSpeechAnalyzer:
return .apple
- case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, .whisperLargeTurbo, .whisperLarge:
+ case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, /* .whisperLargeTurbo, */ .whisperLarge: // buggy - so removed temporarily
return .openai
}
}
@@ -1778,7 +1778,7 @@ final class SettingsStore: ObservableObject {
return "NVIDIA"
case .appleSpeech, .appleSpeechAnalyzer:
return "Apple"
- case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, .whisperLargeTurbo, .whisperLarge:
+ case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, /* .whisperLargeTurbo, */ .whisperLarge: // buggy - so removed temporarily
return "OpenAI"
}
}
@@ -1798,7 +1798,7 @@ final class SettingsStore: ObservableObject {
return "#76B900"
case .appleSpeech, .appleSpeechAnalyzer:
return "#A2AAAD" // Apple Gray
- case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, .whisperLargeTurbo, .whisperLarge:
+ case .whisperTiny, .whisperBase, .whisperSmall, .whisperMedium, /* .whisperLargeTurbo, */ .whisperLarge: // buggy - so removed temporarily
return "#10A37F" // OpenAI Teal
}
}
diff --git a/Sources/Fluid/Services/ASRService.swift b/Sources/Fluid/Services/ASRService.swift
index 17e94a7f..7c22bdeb 100644
--- a/Sources/Fluid/Services/ASRService.swift
+++ b/Sources/Fluid/Services/ASRService.swift
@@ -89,6 +89,7 @@ final class ASRService: ObservableObject {
@Published var isLoadingModel: Bool = false // True when loading cached model into memory (not downloading)
@Published var modelsExistOnDisk: Bool = false
@Published var downloadProgress: Double? = nil
+ @Published var downloadingModelId: String? = nil // Tracks which model is currently being downloaded
private var downloadProgressTask: Task<Void, Never>?
private var hasCompletedFirstTranscription: Bool = false // Track if model has warmed up with first transcription
@@ -190,6 +191,54 @@ final class ASRService: ObservableObject {
SettingsStore.shared.selectedSpeechModel.displayName
}
+ /// Exposes the transcription provider for file transcription (MeetingTranscriptionService)
+ /// This allows file transcription to work with any provider (Parakeet, Whisper, etc.)
+ var fileTranscriptionProvider: TranscriptionProvider {
+ self.transcriptionProvider
+ }
+
+ /// Gets a provider for a specific model (without changing the active selection)
+ /// Used for downloading models without switching the active model.
+ private func getProvider(for model: SettingsStore.SpeechModel) -> TranscriptionProvider {
+ switch model {
+ case .appleSpeechAnalyzer:
+ if #available(macOS 26.0, *) {
+ return AppleSpeechAnalyzerProvider()
+ } else {
+ return AppleSpeechProvider()
+ }
+ case .appleSpeech:
+ return AppleSpeechProvider()
+ case .parakeetTDT, .parakeetTDTv2:
+ // Create a new provider configured for the specific model
+ let provider = FluidAudioProvider(modelOverride: model)
+ return provider
+ default:
+ // Whisper models - create provider with specific model override
+ let provider = WhisperProvider(modelOverride: model)
+ return provider
+ }
+ }
+
+ /// Downloads a specific model without changing the active selection.
+ /// - Parameters:
+ /// - model: The model to download
+ /// - progressHandler: Optional callback for download progress (0.0 to 1.0)
+ func downloadModel(_ model: SettingsStore.SpeechModel, progressHandler: ((Double) -> Void)?) async throws {
+ DebugLogger.shared.info("Downloading model: \(model.displayName) (without changing active selection)", source: "ASRService")
+
+ // Get a fresh provider for this specific model (uses modelOverride for Whisper)
+ let provider = self.getProvider(for: model)
+
+ // Prepare (download) the model
+ try await provider.prepare(progressHandler: { progress in
+ let clamped = max(0.0, min(1.0, progress))
+ progressHandler?(clamped)
+ })
+
+ DebugLogger.shared.info("Model download completed: \(model.displayName)", source: "ASRService")
+ }
+
/// Call this when the transcription provider setting changes to reset state
func resetTranscriptionProvider() {
let newModel = SettingsStore.shared.selectedSpeechModel
@@ -1588,6 +1637,12 @@ final class ASRService: ObservableObject {
/// - Network connectivity issues
/// - Insufficient disk space
func ensureAsrReady() async throws {
+ try await self.ensureAsrReady(progressHandler: nil)
+ }
+
+ /// Ensures ASR models are ready, with an optional external progress handler.
+ /// - Parameter progressHandler: Optional callback for download progress (0.0 to 1.0)
+ func ensureAsrReady(progressHandler: ((Double) -> Void)?) async throws {
let provider = self.transcriptionProvider
let providerKey = "\(type(of: provider)):\(provider.name)"
@@ -1598,7 +1653,7 @@ final class ASRService: ObservableObject {
}
let task = Task { @MainActor in
- try await self.performEnsureAsrReady(provider: provider)
+ try await self.performEnsureAsrReady(provider: provider, externalProgressHandler: progressHandler)
}
self.ensureReadyTask = task
self.ensureReadyProviderKey = providerKey
@@ -1613,7 +1668,7 @@ final class ASRService: ObservableObject {
try await task.value
}
- private func performEnsureAsrReady(provider: TranscriptionProvider) async throws {
+ private func performEnsureAsrReady(provider: TranscriptionProvider, externalProgressHandler: ((Double) -> Void)? = nil) async throws {
// Check if already ready
if self.isAsrReady, provider.isReady {
DebugLogger.shared.debug("ASR already ready with loaded models, skipping initialization", source: "ASRService")
@@ -1681,6 +1736,8 @@ final class ASRService: ObservableObject {
DispatchQueue.main.async {
let clamped = max(0.0, min(1.0, progress))
self?.downloadProgress = clamped
+ // Also call external progress handler if provided
+ externalProgressHandler?(clamped)
}
})
let downloadDuration = Date().timeIntervalSince(downloadStartTime)
diff --git a/Sources/Fluid/Services/FluidAudioProvider.swift b/Sources/Fluid/Services/FluidAudioProvider.swift
index c58d1bd5..0822d526 100644
--- a/Sources/Fluid/Services/FluidAudioProvider.swift
+++ b/Sources/Fluid/Services/FluidAudioProvider.swift
@@ -16,10 +16,18 @@ final class FluidAudioProvider: TranscriptionProvider {
private var asrManager: AsrManager?
private(set) var isReady: Bool = false
+ /// Optional model override - if set, uses this model instead of the global setting.
+ /// Used for downloading specific models without changing the active selection.
+ var modelOverride: SettingsStore.SpeechModel?
+
+ init(modelOverride: SettingsStore.SpeechModel? = nil) {
+ self.modelOverride = modelOverride
+ }
+
func prepare(progressHandler: ((Double) -> Void)? = nil) async throws {
guard self.isReady == false else { return }
- let selectedModel = SettingsStore.shared.selectedSpeechModel
+ let selectedModel = self.modelOverride ?? SettingsStore.shared.selectedSpeechModel
DebugLogger.shared.info("FluidAudioProvider: Starting model preparation for \(selectedModel.displayName)", source: "FluidAudioProvider")
// Download and load models
diff --git a/Sources/Fluid/Services/MeetingTranscriptionService.swift b/Sources/Fluid/Services/MeetingTranscriptionService.swift
index 8bc40f85..4bf864e5 100644
--- a/Sources/Fluid/Services/MeetingTranscriptionService.swift
+++ b/Sources/Fluid/Services/MeetingTranscriptionService.swift
@@ -1,10 +1,7 @@
import AVFoundation
+import Combine
import CoreMedia
import Foundation
-#if arch(arm64)
-import FluidAudio
-#endif
-import Combine
/// Result of a transcription operation
struct TranscriptionResult: Identifiable, Sendable, Codable {
@@ -115,9 +112,10 @@ final class MeetingTranscriptionService: ObservableObject {
try await self.initializeModels()
}
- #if arch(arm64)
- guard let asrManager = asrService.asrManager else {
- throw TranscriptionError.modelLoadFailed("ASR Manager not initialized")
+ // Get the current transcription provider (works for both Parakeet and Whisper)
+ let provider = self.asrService.fileTranscriptionProvider
+ guard provider.isReady else {
+ throw TranscriptionError.modelLoadFailed("Transcription provider not ready")
}
// Check file extension
@@ -149,11 +147,11 @@ final class MeetingTranscriptionService: ObservableObject {
let chunkDurationSeconds: Double = 20 * 60 // 20 minutes per chunk (well under 24min model limit)
let sampleRate: Double = 16_000 // Target sample rate for ASR
let samplesPerChunk = Int(chunkDurationSeconds * sampleRate)
-
+
var allTranscriptions: [String] = []
var totalConfidence: Float = 0
var chunkCount = 0
-
+
// Open audio file for reading
let audioFile: AVAudioFile
do {
@@ -161,34 +159,34 @@ final class MeetingTranscriptionService: ObservableObject {
} catch {
throw TranscriptionError.audioConversionFailed("Could not open audio file: \(error.localizedDescription)")
}
-
+
let fileFormat = audioFile.processingFormat
let totalFrames = AVAudioFrameCount(audioFile.length)
let fileSampleRate = fileFormat.sampleRate
let resampleRatio = sampleRate / fileSampleRate
-
+
// Calculate chunk size in source file frames
let sourceFramesPerChunk = AVAudioFrameCount(Double(samplesPerChunk) / resampleRatio)
var currentFrame: AVAudioFramePosition = 0
-
+
self.currentStatus = duration > 0 ? "Transcribing audio (\(Int(duration))s)..." : "Transcribing audio..."
-
+
while currentFrame < audioFile.length {
let remainingFrames = AVAudioFrameCount(audioFile.length - currentFrame)
let framesToRead = min(sourceFramesPerChunk, remainingFrames)
-
+
// Read chunk from file
guard let buffer = AVAudioPCMBuffer(pcmFormat: fileFormat, frameCapacity: framesToRead) else {
throw TranscriptionError.audioConversionFailed("Could not create audio buffer")
}
-
+
audioFile.framePosition = currentFrame
do {
try audioFile.read(into: buffer, frameCount: framesToRead)
} catch {
throw TranscriptionError.audioConversionFailed("Could not read audio chunk: \(error.localizedDescription)")
}
-
+
// Convert buffer to 16kHz mono Float32 samples
let samples: [Float]
do {
@@ -196,34 +194,34 @@ final class MeetingTranscriptionService: ObservableObject {
} catch {
throw TranscriptionError.audioConversionFailed("Could not resample audio: \(error.localizedDescription)")
}
-
+
// Skip if chunk is too short (< 1 second)
guard samples.count >= Int(sampleRate) else {
currentFrame += AVAudioFramePosition(framesToRead)
continue
}
-
- // Transcribe this chunk
- let chunkResult = try await asrManager.transcribe(samples, source: AudioSource.system)
-
+
+ // Transcribe this chunk using the provider (works for both Parakeet and Whisper)
+ let chunkResult = try await provider.transcribe(samples)
+
if !chunkResult.text.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty {
allTranscriptions.append(chunkResult.text)
totalConfidence += chunkResult.confidence
chunkCount += 1
}
-
+
currentFrame += AVAudioFramePosition(framesToRead)
-
+
// Update progress
let progressPercent = Double(currentFrame) / Double(audioFile.length)
self.progress = 0.3 + (progressPercent * 0.6) // Progress from 30% to 90%
self.currentStatus = "Transcribing... \(Int(progressPercent * 100))%"
}
-
+
// Combine all chunk transcriptions
let finalText = allTranscriptions.joined(separator: " ")
let avgConfidence = chunkCount > 0 ? totalConfidence / Float(chunkCount) : 0
-
+
let transcriptionResult = (text: finalText, confidence: avgConfidence)
self.currentStatus = "Complete!"
@@ -251,10 +249,6 @@ final class MeetingTranscriptionService: ObservableObject {
self.result = result
return result
- #else
- throw TranscriptionError
- .transcriptionFailed("File transcription is only supported on Apple Silicon Macs at this time.")
- #endif
} catch let error as TranscriptionError {
self.error = error.localizedDescription
@@ -316,9 +310,9 @@ final class MeetingTranscriptionService: ObservableObject {
self.currentStatus = ""
self.progress = 0.0
}
-
+
// MARK: - Audio Resampling Helpers
-
+
/// Resample an audio buffer to 16kHz mono Float32 samples
/// - Parameters:
/// - buffer: Source audio buffer
@@ -328,7 +322,7 @@ final class MeetingTranscriptionService: ObservableObject {
let sourceFormat = buffer.format
let sourceSampleRate = sourceFormat.sampleRate
let sourceChannels = sourceFormat.channelCount
-
+
// Create target format (16kHz mono Float32)
guard let targetFormat = AVAudioFormat(
commonFormat: .pcmFormatFloat32,
@@ -336,40 +330,52 @@ final class MeetingTranscriptionService: ObservableObject {
channels: 1,
interleaved: false
) else {
- throw NSError(domain: "MeetingTranscriptionService", code: -1,
- userInfo: [NSLocalizedDescriptionKey: "Could not create target audio format"])
+ throw NSError(
+ domain: "MeetingTranscriptionService",
+ code: -1,
+ userInfo: [NSLocalizedDescriptionKey: "Could not create target audio format"]
+ )
}
-
+
// If already in correct format, just extract samples
- if sourceSampleRate == targetSampleRate && sourceChannels == 1 {
+ if sourceSampleRate == targetSampleRate, sourceChannels == 1 {
guard let channelData = buffer.floatChannelData else {
- throw NSError(domain: "MeetingTranscriptionService", code: -2,
- userInfo: [NSLocalizedDescriptionKey: "Could not access audio channel data"])
+ throw NSError(
+ domain: "MeetingTranscriptionService",
+ code: -2,
+ userInfo: [NSLocalizedDescriptionKey: "Could not access audio channel data"]
+ )
}
let frameLength = Int(buffer.frameLength)
return Array(UnsafeBufferPointer(start: channelData[0], count: frameLength))
}
-
+
// Create converter
guard let converter = AVAudioConverter(from: sourceFormat, to: targetFormat) else {
- throw NSError(domain: "MeetingTranscriptionService", code: -3,
- userInfo: [NSLocalizedDescriptionKey: "Could not create audio converter"])
+ throw NSError(
+ domain: "MeetingTranscriptionService",
+ code: -3,
+ userInfo: [NSLocalizedDescriptionKey: "Could not create audio converter"]
+ )
}
-
+
// Calculate output buffer size
let ratio = targetSampleRate / sourceSampleRate
let estimatedFrameCount = AVAudioFrameCount(Double(buffer.frameLength) * ratio)
-
+
guard let outputBuffer = AVAudioPCMBuffer(pcmFormat: targetFormat, frameCapacity: estimatedFrameCount + 1024) else {
- throw NSError(domain: "MeetingTranscriptionService", code: -4,
- userInfo: [NSLocalizedDescriptionKey: "Could not create output buffer"])
+ throw NSError(
+ domain: "MeetingTranscriptionService",
+ code: -4,
+ userInfo: [NSLocalizedDescriptionKey: "Could not create output buffer"]
+ )
}
-
+
// Convert
var error: NSError?
var inputConsumed = false
-
- converter.convert(to: outputBuffer, error: &error) { inNumPackets, outStatus in
+
+ converter.convert(to: outputBuffer, error: &error) { _, outStatus in
if inputConsumed {
outStatus.pointee = .noDataNow
return nil
@@ -378,17 +384,20 @@ final class MeetingTranscriptionService: ObservableObject {
outStatus.pointee = .haveData
return buffer
}
-
+
if let error = error {
throw error
}
-
+
// Extract samples
guard let channelData = outputBuffer.floatChannelData else {
- throw NSError(domain: "MeetingTranscriptionService", code: -5,
- userInfo: [NSLocalizedDescriptionKey: "Could not access converted audio data"])
+ throw NSError(
+ domain: "MeetingTranscriptionService",
+ code: -5,
+ userInfo: [NSLocalizedDescriptionKey: "Could not access converted audio data"]
+ )
}
-
+
let frameLength = Int(outputBuffer.frameLength)
return Array(UnsafeBufferPointer(start: channelData[0], count: frameLength))
}
diff --git a/Sources/Fluid/Services/WhisperProvider.swift b/Sources/Fluid/Services/WhisperProvider.swift
index 9b6cceee..c22d7c9c 100644
--- a/Sources/Fluid/Services/WhisperProvider.swift
+++ b/Sources/Fluid/Services/WhisperProvider.swift
@@ -19,15 +19,21 @@ final class WhisperProvider: TranscriptionProvider {
private let overriddenModelDirectory: URL?
private let urlSession: URLSession
- init(modelDirectory: URL? = nil, urlSession: URLSession = .shared) {
+ /// Optional model override - if set, uses this model instead of the global setting.
+ /// Used for downloading specific models without changing the active selection.
+ var modelOverride: SettingsStore.SpeechModel?
+
+ init(modelDirectory: URL? = nil, urlSession: URLSession = .shared, modelOverride: SettingsStore.SpeechModel? = nil) {
self.overriddenModelDirectory = modelDirectory
self.urlSession = urlSession
+ self.modelOverride = modelOverride
}
- /// Model filename to use - reads from the unified SpeechModel setting
+ /// Model filename to use - reads from override first, then unified SpeechModel setting
/// Models: tiny (~75MB), base (~142MB), small (~466MB), medium (~1.5GB), large (~2.9GB)
private var modelName: String {
- let configured = SettingsStore.shared.selectedSpeechModel.whisperModelFile?
+ let model = self.modelOverride ?? SettingsStore.shared.selectedSpeechModel
+ let configured = model.whisperModelFile?
.trimmingCharacters(in: .whitespacesAndNewlines)
if let configured, !configured.isEmpty {
return configured
@@ -63,8 +69,8 @@ final class WhisperProvider: TranscriptionProvider {
return 300 * 1024 * 1024
case "ggml-medium.bin":
return 1000 * 1024 * 1024
- case "ggml-large-v3-turbo.bin":
- return 1200 * 1024 * 1024
+ // case "ggml-large-v3-turbo.bin": // buggy - so removed temporarily
+ // return 1200 * 1024 * 1024
case "ggml-large-v3.bin":
return 2000 * 1024 * 1024
default:
diff --git a/Sources/Fluid/UI/AISettings/VoiceEngineSettingsViewModel.swift b/Sources/Fluid/UI/AISettings/VoiceEngineSettingsViewModel.swift
index abe82e01..e25f5531 100644
--- a/Sources/Fluid/UI/AISettings/VoiceEngineSettingsViewModel.swift
+++ b/Sources/Fluid/UI/AISettings/VoiceEngineSettingsViewModel.swift
@@ -20,6 +20,11 @@ final class VoiceEngineSettingsViewModel: ObservableObject {
@Published var suppressSpeechProviderSync: Bool = false
@Published var skipNextSpeechModelSync: Bool = false
+ /// The model currently being downloaded (nil if no download in progress)
+ @Published var downloadingModel: SettingsStore.SpeechModel?
+ /// Download progress for the current model (0.0 to 1.0)
+ @Published var downloadProgress: Double = 0.0
+
@Published var removeFillerWordsEnabled: Bool
init(settings: SettingsStore, appServices: AppServices) {
@@ -107,34 +112,29 @@ final class VoiceEngineSettingsViewModel: ObservableObject {
func downloadSpeechModel(_ model: SettingsStore.SpeechModel) {
guard !self.asr.isRunning else { return }
- let previousActive = self.settings.selectedSpeechModel
+ guard self.downloadingModel == nil else { return } // Prevent concurrent downloads
Task {
- let shouldRestore = previousActive != model
- await MainActor.run {
- if shouldRestore {
- self.suppressSpeechProviderSync = true
- }
- self.settings.selectedSpeechModel = model
- self.asr.resetTranscriptionProvider()
- }
+ // Mark this model as downloading
+ self.downloadingModel = model
+ self.downloadProgress = 0.0
- defer {
- Task { @MainActor in
- guard shouldRestore else { return }
- if self.settings.selectedSpeechModel == model {
- self.skipNextSpeechModelSync = true
- self.settings.selectedSpeechModel = previousActive
- self.asr.resetTranscriptionProvider()
- }
- if self.previewSpeechModel == model {
- self.previewSpeechModel = model
+ do {
+ // Download the model WITHOUT changing the active selection
+ try await self.asr.downloadModel(model, progressHandler: { [weak self] progress in
+ Task { @MainActor in
+ self?.downloadProgress = progress
}
- self.suppressSpeechProviderSync = false
- }
+ })
+
+ DebugLogger.shared.info("Model download completed: \(model.displayName)", source: "VoiceEngineVM")
+ } catch {
+ DebugLogger.shared.error("Failed to download model \(model.displayName): \(error)", source: "VoiceEngineVM")
}
- await self.downloadModels()
+ // Clear downloading state
+ self.downloadingModel = nil
+ self.downloadProgress = 0.0
}
}
diff --git a/Sources/Fluid/UI/AISettingsView+SpeechRecognition.swift b/Sources/Fluid/UI/AISettingsView+SpeechRecognition.swift
index 9e79c97f..54c410cc 100644
--- a/Sources/Fluid/UI/AISettingsView+SpeechRecognition.swift
+++ b/Sources/Fluid/UI/AISettingsView+SpeechRecognition.swift
@@ -276,7 +276,19 @@ extension VoiceEngineSettingsView {
}
.frame(maxWidth: .infinity, alignment: .leading)
- if (self.viewModel.asr.isDownloadingModel || self.viewModel.asr.isLoadingModel) && isActive && !self.viewModel.asr.isAsrReady {
+ // Action area: Show progress if THIS model is being downloaded
+ if self.viewModel.downloadingModel == model {
+ // This specific model is currently being downloaded
+ VStack(alignment: .trailing, spacing: 4) {
+ ProgressView(value: self.viewModel.downloadProgress)
+ .progressViewStyle(.linear)
+ .frame(width: 90)
+ Text("\(Int(self.viewModel.downloadProgress * 100))%")
+ .font(.caption2)
+ .foregroundStyle(.secondary)
+ }
+ } else if (self.viewModel.asr.isDownloadingModel || self.viewModel.asr.isLoadingModel) && isActive && !self.viewModel.asr.isAsrReady {
+ // Active model is loading/downloading (for Activate flow)
VStack(alignment: .trailing, spacing: 4) {
if let progress = self.viewModel.asr.downloadProgress, self.viewModel.asr.isDownloadingModel {
ProgressView(value: progress)
@@ -313,7 +325,7 @@ extension VoiceEngineSettingsView {
.tint(Color.fluidGreen)
.fontWeight(.semibold)
.shadow(color: Color.fluidGreen.opacity(0.35), radius: 4, x: 0, y: 1)
- .disabled(self.viewModel.asr.isRunning)
+ .disabled(self.viewModel.asr.isRunning || self.viewModel.downloadingModel != nil)
}
if !model.usesAppleLogo {
@@ -326,7 +338,7 @@ extension VoiceEngineSettingsView {
.foregroundStyle(.red.opacity(0.7))
}
.buttonStyle(.plain)
- .disabled(self.viewModel.asr.isRunning)
+ .disabled(self.viewModel.asr.isRunning || self.viewModel.downloadingModel != nil)
.offset(x: isSelected ? 0 : 12)
.opacity(isSelected ? 1 : 0)
}
@@ -346,7 +358,7 @@ extension VoiceEngineSettingsView {
.buttonStyle(.borderedProminent)
.controlSize(.small)
.tint(.blue)
- .disabled(self.viewModel.asr.isRunning)
+ .disabled(self.viewModel.asr.isRunning || self.viewModel.downloadingModel != nil)
.offset(x: isSelected ? 0 : 16)
.opacity(isSelected ? 1 : 0)
}