v0.5.3

What's New in v0.5.2

Grid3 & System.Speech Compatibility

• Cloud voices only via registry promotion — Azure/Edge voices are no longer enumerated dynamically. Only voices promoted via "Install Selected" (HKLM registry tokens) appear in SAPI. This fixes Grid3 and other System.Speech apps that can't activate in-memory tokens
• Voice promotion fixed — PromoteElevated was silently failing because EngineConfig.exe was removed. Now uses reg.exe import with UAC elevation, creating tokens in both legacy SAPI and Speech_OneCore paths
• Voice enumerator deadlock fixed — double mutex lock in VoiceTokenEnumerator::EnumTokens caused "resource deadlock would occur" crash on first enumeration or after 10s cache expiry
• No more voice flood — previously 500+ Azure voices flooded SAPI voice lists; now only promoted voices appear

SSML & Speech Markdown via Web Speech API

• PUA sentinel — web apps using Chrome/Firefox Web Speech API can embed SSML or Speech Markdown via Unicode Private Use Area characters:
  • \uE000\uE001\uE002 prefix = SSML mode
  • \uE000\uE001\uE003 prefix = Speech Markdown mode
  • Other SAPI voices ignore the PUA characters harmlessly
• SSML normalization — adapter automatically injects required <speak version="1.0" xmlns="..." xml:lang="..."> attributes when missing
• Speech Markdown syntax — use (text)[rate:"slow"] format (standard Speech Markdown spec)

Chrome & Edge Support

• Speech_OneCore registration — VoiceGarden voices now appear in Chrome and Edge, not just legacy SAPI applications
• Voices registered in both SOFTWARE\Microsoft\Speech\Voices\Tokens (Firefox, Grid3, Balabolka) and SOFTWARE\Microsoft\Speech_OneCore\Voices\Tokens (Chrome, Edge) when promoted

RustTtsWrapper 0.3.14

• SSML pass-through fixed — tts_speak_ssml() correctly detects and processes <speak> documents
• Speech Markdown converter fixed — produces valid SSML with required Azure attributes
• Voice selection fixed — tts_set_voice() correctly selects the requested Azure voice
• GetVoices crash fixed — cloud engine voice listing no longer crashes with access violation

Voice Engine Fixes

• Azure voice activation — Azure voice tokens now include EngineType in VoiceGardenConfig
• Cloud engines off by default — paid cloud engines disabled on fresh install; SherpaOnnx (offline) and Edge (free) remain on

MMS Model Data Fix

• 1143 MMS models now searchable — language metadata was blank due to mismatched JSON field names. Kurdish, Aceh, Abidji, and 1100+ other languages now appear correctly

Localization (36 Languages)

• Full UI translation — all user-facing strings in Strings.resx
• 36 language packs — Arabic, Bengali, Chinese, Czech, Danish, Dutch, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Korean, Malay, Persian, Polish, Portuguese, Punjabi, Russian, Spanish, Swahili, Swedish, Tamil, Telugu, Thai, Turkish, Ukrainian, Urdu, Vietnamese, and more
• Automatic language detection — app language follows OS locale
• RTL support — Arabic, Hebrew, Urdu, and Persian layouts mirror automatically

Credential Management

• Auto-save credentials — API keys saved to registry immediately
• Dual-screen sync — credentials shared between Configure Credentials and Configure Voices

Installer Fixes

• Error 1721/1722 — Return="ignore" on regsvr32 prevents install/uninstall failure
• Upgrade ghost fix — IgnoreRemoveFailure + Schedule="afterInstallFinalize" allows clean installs over broken previous versions

Onboarding

• 3-page wizard — Welcome, Getting Started, Privacy (analytics opt-in)
• Version-gated — bump CurrentOnboardingVersion to force all users to see updated content

---

What's New in v0.4.0

Major Architecture Changes

• RustTtsWrapper Integration — complete transition to Rust-based TTS wrapper
  • Removed legacy DotNetTtsWrapper and EngineConfig components
  • Full x86 and x64 support with native rust_tts_wrapper.dll

Breaking Changes

• Removed Components — DotNetTtsWrapper and EngineConfig no longer supported
• Minimum Requirements — Windows 10+