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

Added Parakeet V2 and readme changes for new model addtions

altic-dev6534f97154fb
diff --git a/README.md b/README.md
index 1b053458..6523d92f 100644
--- a/README.md
+++ b/README.md
@@ -1,5 +1,7 @@
# FluidVoice
+[![Supported Models](https://img.shields.io/badge/Models-Parakeet%20v3%20%26%20v2%20%7C%20Apple%20Speech%20%7C%20Whisper-blue)](https://github.com/altic-dev/Fluid-oss)
+
Fully open source voice-to-text dictation app for macOS with AI enhancement.
**Get the latest release from [here](https://github.com/altic-dev/Fluid-oss/releases/latest)**
@@ -44,16 +46,32 @@ https://github.com/user-attachments/assets/c57ef6d5-f0a1-4a3f-a121-637533442c24
- **New History stats**
- **Stats to monitor usage**
+
## Features
- **Live Preview Mode**: Real-time transcription preview in overlay
-- **Real-time transcription** using Parakeet TDT v3 model
+- **Multiple Speech Models**: Parakeet TDT v3 & v2, Apple Speech, and Whisper
+- **Real-time transcription** with extremely low latency
- **AI enhancement** with OpenAI, Groq, and custom providers
-- **25+ languages** with auto-detection
- **Global hotkey** for instant voice capture
- **Smart typing** directly into any app
- **Menu bar integration** for quick access
- **Auto-updates** with seamless restart
+## Supported Models
+
+### Parakeet TDT v3 (Default)
+Optimized for Apple Silicon. Supports 25 languages with auto-detection:
+**Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Russian, Ukrainian.**
+
+### Parakeet TDT v2
+English-only model optimized for higher accuracy and consistency on Apple Silicon.
+
+### Apple Speech
+Uses the native macOS Speech Recognition engine and Speech Analyzer API (macOS 26+).
+
+### Whisper
+Universal support (runs on Intel & Apple Silicon). Supports 99 languages.
+
## Quick Start
1. Download the latest release
diff --git a/Sources/Fluid/Persistence/SettingsStore.swift b/Sources/Fluid/Persistence/SettingsStore.swift
index 30b3ec8b..25cb08d4 100644
--- a/Sources/Fluid/Persistence/SettingsStore.swift
+++ b/Sources/Fluid/Persistence/SettingsStore.swift
@@ -963,6 +963,7 @@ final class SettingsStore: ObservableObject {
// MARK: - FluidAudio Models (Apple Silicon Only)
case parakeetTDT = "parakeet-tdt"
+ case parakeetTDTv2 = "parakeet-tdt-v2"
// MARK: - Apple Native
@@ -984,7 +985,8 @@ final class SettingsStore: ObservableObject {
var displayName: String {
switch self {
- case .parakeetTDT: return "Parakeet TDT"
+ case .parakeetTDT: return "Parakeet TDT v3 (Multilingual)"
+ case .parakeetTDTv2: return "Parakeet TDT v2 (English Only)"
case .appleSpeech: return "Apple Speech (Legacy)"
case .appleSpeechAnalyzer: return "Apple Speech (macOS 26+)"
case .whisperTiny: return "Whisper Tiny"
@@ -999,6 +1001,7 @@ final class SettingsStore: ObservableObject {
var languageSupport: String {
switch self {
case .parakeetTDT: return "25 Languages"
+ 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:
@@ -1009,6 +1012,7 @@ final class SettingsStore: ObservableObject {
var downloadSize: String {
switch self {
case .parakeetTDT: return "~500 MB"
+ case .parakeetTDTv2: return "~500 MB"
case .appleSpeech: return "Built-in (Zero Download)"
case .appleSpeechAnalyzer: return "Built-in"
case .whisperTiny: return "~75 MB"
@@ -1022,14 +1026,14 @@ final class SettingsStore: ObservableObject {
var requiresAppleSilicon: Bool {
switch self {
- case .parakeetTDT: return true
+ case .parakeetTDT, .parakeetTDTv2: return true
default: return false
}
}
var isWhisperModel: Bool {
switch self {
- case .parakeetTDT, .appleSpeech, .appleSpeechAnalyzer: return false
+ case .parakeetTDT, .parakeetTDTv2, .appleSpeech, .appleSpeechAnalyzer: return false
default: return true
}
}
diff --git a/Sources/Fluid/Services/ASRService.swift b/Sources/Fluid/Services/ASRService.swift
index d72e4953..e811700b 100644
--- a/Sources/Fluid/Services/ASRService.swift
+++ b/Sources/Fluid/Services/ASRService.swift
@@ -61,12 +61,16 @@ private actor TranscriptionExecutor {
/// ```
///
/// ## Language Support
-/// The service automatically detects and transcribes 25 European languages with Parakeet TDT v3:
-/// Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German,
-/// Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian,
-/// Slovak, Slovenian, Spanish, Swedish, Russian, and Ukrainian.
+/// The service supports multiple models with varying language capabilities:
+/// - **Parakeet TDT v3** (Default): Automatically detects and transcribes 25 European languages:
+/// Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German,
+/// Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian,
+/// Slovak, Slovenian, Spanish, Swedish, Russian, and Ukrainian.
+/// - **Parakeet TDT v2**: Specialized for high-accuracy English transcription.
+/// - **Apple Speech**: Supports all system languages available on macOS.
+/// - **Whisper**: Supports 99 languages.
///
-/// No manual language selection is required - the model automatically detects the spoken language.
+/// No manual language selection is required for Parakeet models - v3 automatically detects the spoken language.
/// ## Thread Safety
/// All public methods are marked with @MainActor to ensure thread safety.
/// Audio processing happens on background threads for optimal performance.
@@ -129,7 +133,7 @@ final class ASRService: ObservableObject {
}
case .appleSpeech:
return self.getAppleSpeechProvider()
- case .parakeetTDT:
+ case .parakeetTDT, .parakeetTDTv2:
return self.getFluidAudioProvider()
default:
return self.getWhisperProvider()
@@ -193,6 +197,12 @@ final class ASRService: ObservableObject {
self.ensureReadyTask = nil
self.ensureReadyProviderKey = nil
+ // Reset cached providers to force re-initialization with new settings
+ self.fluidAudioProvider = nil
+ self.whisperProvider = nil
+ self.appleSpeechProvider = nil
+ self._appleSpeechAnalyzerProvider = nil
+
// CRITICAL FIX: Check if the NEW model's files exist on disk
// This prevents UI from showing "Download" when model is already downloaded
// Use Task for async check to support providers like AppleSpeechAnalyzerProvider
diff --git a/Sources/Fluid/Services/FluidAudioProvider.swift b/Sources/Fluid/Services/FluidAudioProvider.swift
index 7bd8c9ce..c58d1bd5 100644
--- a/Sources/Fluid/Services/FluidAudioProvider.swift
+++ b/Sources/Fluid/Services/FluidAudioProvider.swift
@@ -19,10 +19,18 @@ final class FluidAudioProvider: TranscriptionProvider {
func prepare(progressHandler: ((Double) -> Void)? = nil) async throws {
guard self.isReady == false else { return }
- DebugLogger.shared.info("FluidAudioProvider: Starting model preparation", source: "FluidAudioProvider")
+ let selectedModel = SettingsStore.shared.selectedSpeechModel
+ DebugLogger.shared.info("FluidAudioProvider: Starting model preparation for \(selectedModel.displayName)", source: "FluidAudioProvider")
// Download and load models
- let models = try await AsrModels.downloadAndLoad()
+ let models: AsrModels
+ if selectedModel == .parakeetTDTv2 {
+ // Explicitly load v2 (English Only)
+ models = try await AsrModels.downloadAndLoad(version: .v2)
+ } else {
+ // Default to v3 (Multilingual)
+ models = try await AsrModels.downloadAndLoad(version: .v3)
+ }
// Initialize AsrManager
let manager = AsrManager(config: ASRConfig.default)
@@ -48,23 +56,34 @@ final class FluidAudioProvider: TranscriptionProvider {
func modelsExistOnDisk() -> Bool {
let baseCacheDir = AsrModels.defaultCacheDirectory().deletingLastPathComponent()
- let v3CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v3-coreml")
- return FileManager.default.fileExists(atPath: v3CacheDir.path)
+
+ if SettingsStore.shared.selectedSpeechModel == .parakeetTDTv2 {
+ let v2CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v2-coreml")
+ return FileManager.default.fileExists(atPath: v2CacheDir.path)
+ } else {
+ let v3CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v3-coreml")
+ return FileManager.default.fileExists(atPath: v3CacheDir.path)
+ }
}
func clearCache() async throws {
let baseCacheDir = AsrModels.defaultCacheDirectory().deletingLastPathComponent()
-
- // Clear v2 cache
- let v2CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v2-coreml")
- if FileManager.default.fileExists(atPath: v2CacheDir.path) {
- try FileManager.default.removeItem(at: v2CacheDir)
- }
-
- // Clear v3 cache
- let v3CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v3-coreml")
- if FileManager.default.fileExists(atPath: v3CacheDir.path) {
- try FileManager.default.removeItem(at: v3CacheDir)
+ let selectedModel = SettingsStore.shared.selectedSpeechModel
+
+ if selectedModel == .parakeetTDTv2 {
+ // Clear v2 cache only
+ let v2CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v2-coreml")
+ if FileManager.default.fileExists(atPath: v2CacheDir.path) {
+ try FileManager.default.removeItem(at: v2CacheDir)
+ DebugLogger.shared.info("FluidAudioProvider: Deleted Parakeet v2 cache", source: "FluidAudioProvider")
+ }
+ } else {
+ // Clear v3 cache only (default)
+ let v3CacheDir = baseCacheDir.appendingPathComponent("parakeet-tdt-0.6b-v3-coreml")
+ if FileManager.default.fileExists(atPath: v3CacheDir.path) {
+ try FileManager.default.removeItem(at: v3CacheDir)
+ DebugLogger.shared.info("FluidAudioProvider: Deleted Parakeet v3 cache", source: "FluidAudioProvider")
+ }
}
self.isReady = false
diff --git a/Sources/Fluid/UI/AISettingsView.swift b/Sources/Fluid/UI/AISettingsView.swift
index d925d815..c4c7e96a 100644
--- a/Sources/Fluid/UI/AISettingsView.swift
+++ b/Sources/Fluid/UI/AISettingsView.swift
@@ -235,7 +235,7 @@ struct AISettingsView: View {
Spacer()
- if !SettingsStore.shared.selectedSpeechModel.isWhisperModel {
+ if SettingsStore.shared.selectedSpeechModel.requiresAppleSilicon {
Text("Apple Silicon")
.font(.caption)
.padding(.horizontal, 8)
@@ -278,7 +278,9 @@ struct AISettingsView: View {
case .appleSpeechAnalyzer:
return "Apple Speech uses advanced on-device recognition with fast, accurate transcription. Requires macOS 26+."
case .parakeetTDT:
- return "Parakeet TDT uses CoreML and Neural Engine for fastest transcription (25 languages) on Apple Silicon."
+ return "Parakeet TDT v3 uses CoreML and Neural Engine for fastest transcription (25 languages) on Apple Silicon."
+ case .parakeetTDTv2:
+ return "Parakeet TDT v2 is an English-only model optimized for accuracy and consistency on Apple Silicon."
default:
return "Whisper models support 99 languages and work on any Mac."
}