New Speech-to-Text Models
- NewAdded higher-accuracy offline recognition models: FireRedASR for the best Chinese–English mixed-speech accuracy, an updated Chinese streaming model (Zipformer 2025), and SenseVoice 2025 for improved 5-language recognition (Chinese, English, Japanese, Korean, Cantonese)
- Superseded model files are cleaned up automatically to free disk space
Automatic Updates
- NewSub!t now checks for new versions in the background and downloads them automatically, then asks whether to restart and install — the prompt is deferred while you are running timecode or output so it never interrupts a show
- A "Check for Updates" menu item lets you check any time
- Updates only replace the app itself; your license, preferences, and downloaded models are always kept
Per-Word Font Size
- NewResize individual words or whole lines independently in both Document Window and SRT Window edit modes
- Toolbar A− / A+ buttons or keyboard shortcuts ⌘↑ / ⌘↓ adjust the selected text in 10% steps
- Relative sizing from 50% to 300% of the base font size — scales proportionally for any output resolution, with no layout drift
- Returning to 100% removes the marker automatically; uses a custom
<sz> tag preserved on save
Paragraph Group Output (TXT Documents Only)
- NewFor plain-text document subtitles only (Document Window, not SRT/timecode cues) — a "Group" toggle segments the text by blank lines, so selecting any line outputs the whole paragraph at once, ideal for lyrics and verses
- Up / Down arrows move paragraph by paragraph, and the preview highlights the entire block that will be sent
- The toggle state is saved with the project; new projects default to off
Improvements & Fixes
- FixPreserve multiple and leading spaces in subtitle output (including NDI)
- FixResolve a dialog overlap that could appear when opening or creating a project
Three-Layer Subtitle Background System
- NewThree independent background layers that stack from back to front for professional subtitle compositing
- Layer 1 — Canvas Background: full-screen output background, supports transparent for NDI output
- Layer 2 — Subtitle Background: background bar behind the subtitle text line with adjustable opacity
- Layer 3 — Per-Word Highlight: background color applied to individual characters via the edit mode toolbar highlighter
- All three layers can be freely combined for professional NDI compositing results
Text Style Toolbar
- NewText style toolbar in Document Window edit mode — apply color, highlighter, bold, and italic to selected text
- NewText style toolbar in SRT Subtitle Window edit mode — same formatting capabilities for SRT subtitle text
- 4 preset text colors (Red, Yellow, Blue, Green) + custom color picker
- 4 preset highlighter colors (Red, Yellow, Blue, Green) + custom color picker for per-word background (Layer 3)
- Clear button to remove all formatting (text color, background color, bold, italic) at once
- Keyboard shortcuts ⌘B / ⌘I for bold and italic
- Text color and bold/italic use SRT standard tags (
<font color> / <b> / <i>); background color uses custom <bg color> tag
- Applied styles preserved when saving and display correctly in subtitle output
LTC Timecode Sync
- NewSMPTE 12M Linear Timecode (LTC) synchronization for frame-accurate subtitle cueing
- Native C++ CoreAudio decoder (subit-ltc-native) for real-time external LTC input
- JavaScript fallback mode with full filter chain for broad compatibility
- LTC audio file playback for offline timecode sync and rehearsal
- Flywheel dropout protection: extrapolates playhead from last known good TC
- Lock-in validation: 4 consecutive incrementing frames required before accepting TC baseline
- TC offset compensation for flexible cue timing adjustment
Auto GO
- NewAuto-advance GO to next file tab when timecode enters its range
- Sequential fallback: when TC range has no match, advances to the next tab with cues
Audio Monitoring
- NewLTC file audio monitoring with AudioWorklet and ring buffer pre-fill
- Drift-compensating playback for accurate sync during file mode
Speech-to-Text (STT)
- NewTriple-engine speech recognition: Deepgram online streaming, Whisper offline, Sherpa-onnx offline streaming / non-streaming
- Deepgram Nova-3 real-time streaming with ~200ms latency, Keywords Boosting support
- Whisper large-v3-turbo offline recognition with Metal GPU acceleration, 99 languages
- Sherpa-onnx streaming mode (Zipformer Chinese-English) and non-streaming mode (SenseVoice 5 languages)
- Audio input device selector, recognition language selection, model management panel
- OpenCC auto Traditional/Simplified Chinese conversion for offline engines
Usability Improvements
- ImprovedEnter / Esc work globally during STT without clicking the input box first
- Improved← → arrow keys to switch document tabs
NDI Output Optimization
- ImprovedCustom native C++ NDI module, performance boosted to 60fps
- Event-driven + heartbeat hybrid mode, idle CPU usage < 1%
Bug Fixes
- FixedTraditional/Simplified Chinese conversion causing incorrect characters
NDI Video Output
- NewNDI LAN broadcast output, accessible by any device on the same network
- BGRA video stream with Alpha Channel transparency support
- Compatible with OBS Studio, Resolume Arena, vMix and more
Subtitle Transitions
- NewFade in/out transition effects with customizable duration (0.0 – 3.0 seconds)
- Independent transition timing per screen
Internationalization
- NewFull trilingual interface: Traditional Chinese, Simplified Chinese, English
- Menu language switching (restart required), automatic system language detection
Styling System
- ImprovedVertical and horizontal positions fully independent, consistent behavior for horizontal/vertical text
- NewFont picker (recommended + system fonts), bold/italic toggle, letter spacing control
- Custom font size memory
New Features
- NewBackground color dropdown (Black, White, Red, Green, Blue, Custom)
- NewAuto-hide subtitles (3s / 5s / 10s)
- Remember last used folder path when opening/saving
Bug Fixes
- FixedDouble checkmarks appearing after language switch
Initial Release
- Real-time subtitle input and display, single/multi-line modes (1–5 lines)
- Multi-screen independent output with per-screen styling and resolution
- Automatic Traditional/Simplified Chinese conversion
- Virtual output window with live preview
- Document management: open .txt / .docx, line-by-line sending, multi-tab, edit mode
- .sub project format, saves all settings and state
- RTL text direction support (Arabic, Hebrew)
- Vertical text dual-side display
