$ show altic-dev/FluidVoice
FluidVoice is a macOS desktop application for real-time voice dictation, meeting transcription, and AI-powered text enhancement. It provides a local-first dictation experience with on-device speech recognition (Parakeet, Nemotron, Whisper, Apple Speech) and optional cloud AI post-processing. The application features a command mode for AI chat, meeting recording with speaker diarization and acoustic echo cancellation, and a local HTTP API for external integrations. The architecture is a native SwiftUI/AppKit application with a service-oriented design. Audio capture uses CoreAudio and ScreenCaptureKit, with a vendored WebRTC audio processing stack for echo cancellation (AEC3), noise suppression, and voice activity detection. Speech recognition is handled by multiple providers, with model downloads from Hugging Face. The application includes a durable PCM storage system with chunk ledger for meeting audio, a signed model release feed for secure updates, and a comprehensive settings/persistence layer using SQLite and UserDefaults. Key entry points include FluidApp.swift (main application), LocalAPIRouter.swift (HTTP API server), MeetingSummaryProvider.swift (AI meeting summaries), ASRService.swift (speech recognition orchestration), and DictationAppSession.swift (dictation workflow). External dependencies include macOS system APIs, SQLite, vendored WebRTC audio processing, and various AI provider APIs (OpenAI, Anthropic, etc.). The repository also contains research-grade audio processing code (WebRTC AEC3, AGC2, RNN VAD) and pronunciation learning evaluation tools, but the dominant character is a production macOS application with extensive testing and CI/CD.
The LLM identified 5 files worth highlighting for reuse:
Implements a durable write-ahead log for meeting audio chunks using intent/checkpoint/terminal files with fsync-before-rename and directory fsync. Provides no-overwrite semantics, path-traversal rejection, and atomic publication via renameatx_np(RENAME_EXCL). This pattern is reusable for any application requiring crash-safe, append-only storage with verification.
Manages exclusive admission, suspend, and restore of large ML models (e.g., diarization, ASR) with TaskLocal grants and background-operation wrappers. Provides veto/termination handling and snapshot-suspend-restore lifecycle. This pattern is reusable for any system needing coordinated access to exclusive resources with cancellation safety.
Parses and verifies a signed, versioned model-release feed with schema validation, URL/SHA-256 checks, and signature verification using CryptoKit. Designed to be dependency-light (Foundation + CryptoKit) so it can be compiled as a standalone tool. Reusable for any application needing secure, verifiable model or asset distribution.
Provides a C API for low-latency CoreAudio capture with lock-free ring-buffer consumer semantics, format-dirty gating, and non-allocating topology trace events. The design is reusable for any audio capture subsystem requiring deterministic performance and minimal allocation.
Implements serialized hardware access with cancellation, timeout, and recovery. Uses NSLock for thread safety and supports one-waiter semantics with stale-session rejection. This pattern is reusable for any system that needs to serialize access to a shared hardware resource with robust error handling.