Skip to content

OmniVox v0.1.8

Choose a tag to compare

View all tags
@trigga6006 trigga6006 released this 30 Mar 00:35
· 84 commits to main since this release
v0.1.8
66e82b0

What's New in v0.1.8

New Features

Vocabulary System

  • Add custom words that Whisper should recognize (names, brands, technical terms) without needing a dictionary replacement
  • Vocabulary words bias the Whisper decoder directly via the initial prompt
  • Full CRUD with per-mode support — each context mode can have its own vocabulary
  • New "Vocabulary" tab on the Dictionary page

Audio Ducking Controls

  • Toggle to enable/disable system volume lowering during dictation
  • Slider (0–100%) to control exactly how much other audio is reduced while recording
  • Previously hardcoded at 70% reduction with no way to disable

Hot-Swap Whisper Prompt

  • Adding or removing dictionary entries and vocabulary words now updates the Whisper decoder prompt instantly — no model reload required
  • Changes take effect on the very next transcription

Improvements

  • Removed dead AI cleanup code (commented-out LLM block)
  • Removed unused llm_prompt field from context mode API
  • Cleaned up verbose debug logging for production
  • Fixed settings page animation staggering
  • Synced TypeScript types with Rust backend (active_context_mode_id)
  • Larger stack allocation for debug builds (64MB main thread, 256MB model loader)
  • Single-instance plugin disabled in debug builds for easier development

Bug Fixes

  • Fixed GPU acceleration being force-disabled in debug builds — dev server now respects the GPU acceleration setting, restoring fast transcription speeds

Full Changelog: v0.1.7...v0.1.8

Assets 3
Loading