SayItFlow v1.2.0

SayItFlow v1.2.0 — Performance & Memory Optimization Release

This release brings dramatic memory and CPU efficiency improvements, transforming SayItFlow into a lightweight background menu bar app that minimizes system load on Apple Silicon Macs.

What's New & Optimized

• 🧠 Intelligent Model Lifecycle Management:
  • Voz Core ML Model: Implemented lazy loading and automatic 120-second idle unloading. After 2 minutes of silence, the 467 MB model weights and Neural Engine residency are cleanly released.
  • S1-mini Normalizer Server: Moved from permanent background execution to on-demand lifecycle with a 120-second inactivity shutdown timer, reclaiming 600–800 MB of GPU/system RSS when idle.
  • Zero Idle Model Overhead: Combined idle memory reduced by ~1.0–1.3 GB, dropping idle RAM usage to just ~15–30 MB.
• ⚡ Accelerate Framework Audio Pipeline:
  • Audio peak level metering replaced with hardware-accelerated vector SIMD math via Apple's Accelerate framework (vDSP_rmsqv).
  • Pre-allocated sample collector capacity (~120s audio buffer) to eliminate runtime dynamic array reallocations during speech.
  • Throttled level monitor polling from 25 Hz to 10 Hz, eliminating unnecessary main-thread context switching.
• 📝 Logging & System Efficiency:
  • Pre-compiled disfluency and speech normalization regexes.
  • Shared ISO-8601 date formatter and persistent FileHandle caching to eliminate filesystem descriptor churn on every log entry.
  • Automatic log file rotation (1 MB threshold with .prev rollover).
• 🖥️ Window & UI Memory Reclamation:
  • Floating HUD NSPanel and SwiftUI hosting view hierarchy are explicitly released upon dismiss to free memory between dictations.

Verification

• All automated test suites passing (make test).
• Ad-hoc signed arm64 DMG installer.

SHA-256 (SayItFlow.dmg): 598a8b9cd8e075aad6101863bf17ba5f657dc4a16149c29382ae5c2ce4cc2ad0

SayItFlow v1.0.0

Initial public release of SayItFlow.\n\n- Native push-to-talk dictation for macOS\n- On-device Desert Ant Voz transcription and S1-mini normalization\n- Apple Silicon (arm64) DMG\n\nSHA-256 (SayItFlow.dmg): 74c40872383af2f4b277af3d0dfdb01e76e46bbbec6c61f0a6b5aed61db3c87d\n\nNote: this build is ad-hoc signed and has not been notarized.