Does FluidVoice Support Real-Time Speech Synthesis?

FluidVoice does not support real-time speech synthesis; it is built exclusively for real-time speech-to-text dictation and transcription.

FluidVoice is an open-source macOS voice dictation application focused entirely on converting spoken audio into written text. While it delivers robust real-time speech recognition through native macOS APIs, the codebase contains no text-to-speech (TTS) engine or speech synthesis functionality.

How FluidVoice Handles Speech Processing

The application architecture centers on three core services that manage live audio capture and streaming transcription. According to the altic-dev/FluidVoice source code, these components work together to deliver low-latency dictation without any synthesis capability.

ASRService.swift: The Central Recognition Pipeline

The ASRService class in Sources/Fluid/Services/ASRService.swift orchestrates the entire speech-to-text workflow. It manages audio buffering, session lifecycle, and delegates recognition tasks to platform-specific providers.

// Start a real-time transcription session
let asr = ASRService.shared
asr.startSession(language: .englishUS) { result in
    switch result {
    case .partial(let text):
        // Update UI with interim transcription
        print("Partial: \(text)")
    case .final(let text):
        // Finalized sentence
        print("Final: \(text)")
    case .error(let err):
        print("Error: \(err)")
    }
}

This callback-based API returns partial and final transcription results—never synthesized audio.

AppleSpeechProvider.swift: macOS Recognition Wrapper

Sources/Fluid/Services/AppleSpeechProvider.swift wraps macOS's built-in SFSpeechRecognizer for standard recognition tasks. As implemented in altic-dev/FluidVoice, this provider only handles incoming audio streams:

// AppleSpeechProvider – low-level call to macOS recognizer
let provider = AppleSpeechProvider()
provider.recognizeAudioStream(url: audioURL) { transcript in
    // transcript contains the spoken words – no synthesis involved
    print(transcript)
}

AppleSpeechAnalyzerProvider.swift: Streaming for Low Latency

For advanced use cases requiring faster partial results, Sources/Fluid/Services/AppleSpeechAnalyzerProvider.swift implements SFSpeechAnalyzer—Apple's streaming recognition API introduced for continuous dictation scenarios. This remains transcription-only with no synthesis path.

Evidence Against Speech Synthesis Support

A comprehensive review of the FluidVoice Swift source tree reveals no TTS implementation:

• No AVSpeechSynthesizer imports or usage — Apple's standard synthesis framework is entirely absent
• No "synthesize" method calls related to audio — the only occurrences reference UI progress bar updates
• No synthesis settings in SettingsStore.swift — located at Sources/Fluid/Persistence/SettingsStore.swift, this file only stores recognition model preferences (appleSpeech, appleSpeechAnalyzer)
• No synthesis UI in ContentView.swift — the main interface at Sources/Fluid/ContentView.swift displays transcription results and recording controls exclusively

What FluidVoice Actually Does

Rather than speech synthesis, FluidVoice provides these real-time speech-to-text features:

• Live streaming transcription with partial and final result callbacks
• Speaker diarization to identify different speakers in recorded audio
• Multiple recognition backends via appleSpeech and appleSpeechAnalyzer models
• Continuous dictation sessions with automatic segmenting and punctuation

Summary

• FluidVoice is speech-to-text only — it converts audio to text, not text to audio
• The codebase contains zero synthesis infrastructure — no AVSpeechSynthesizer, custom TTS engines, or audio generation pipelines
• All speech-related services (ASRService, AppleSpeechProvider, AppleSpeechAnalyzerProvider) handle recognition exclusively
• Users seeking real-time speech synthesis need separate macOS TTS tools or alternative applications

Frequently Asked Questions

Can FluidVoice read text aloud?

No. FluidVoice has no text-to-speech capability. The application only captures microphone input and returns transcribed text. For macOS text-to-speech, use system features like Spoken Content in Accessibility settings or third-party TTS applications.

Does FluidVoice plan to add speech synthesis?

The current codebase shows no foundation for synthesis features. The architecture is purpose-built for recognition efficiency, with ASRService designed around transcription provider patterns that don't accommodate audio output generation.

What macOS speech APIs does FluidVoice use?

FluidVoice uses SFSpeechRecognizer (via AppleSpeechProvider) and SFSpeechAnalyzer (via AppleSpeechAnalyzerProvider)—both Apple frameworks for speech recognition only. These APIs handle audio input and return text transcripts; they do not perform synthesis.

Is there any speech generation in FluidVoice's audio pipeline?

No. The audio pipeline in ASRService.swift is unidirectional: microphone → buffer → recognition provider → text callback. No audio output path exists, and no synthesis classes are referenced anywhere in Sources/Fluid/.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}