Changelog

✦ OS volume slider on every bus

• Where: a small "OS xx%" slider appears below the device picker whenever the selected output has a macOS volume control — yellow while below 100%.
• Ride it or clear it: drag to set the device volume right from the strip; double-click snaps to 100%. Changes made elsewhere (System Settings, the knob on a Bluetooth speaker) show up within seconds.
• When the device is your destination (headphones, a Bluetooth speaker): the OS stage is a perfectly good listening control — now with visible context.
• When the signal travels on (line-out into a desk, an interface, another app): keep it at 100% = bit-transparent, and do your riding on the bus fader. The slider makes sure you see when that stage isn't clean.
• No slider? The device only has a hardware knob (many USB DACs) — there is no macOS stage to worry about. macOS only for now.

✦ What changed

• SAVE AS… button in the toolbar plus Cmd/Ctrl+S (save) and Shift+Cmd/Ctrl+S (save as) — keep separate presets per show, studio or stream.
• Missing devices don't kill the preset anymore: everything loads with all settings, a toast names the absent device, and the moment it comes back the channel or bus reconnects automatically.
• Re-plugged into another port? Device IDs can change with the port — presets now also store the device name and re-match by it.
• No silent fallback to the default output: a missing device shows as "⚠ device missing" instead of visually snapping to default — which could route a bus straight into its own capture source (feedback loop).
• New "(no output)" option: a bus can deliberately run without an output device — for processing or standby buses.

✦ The special case, explained

• Some USB DACs advertise 16-bit modes alongside their 32-bit/float modes. The audio backend preferred 16-bit internally on such devices — and its converter then pulled audio at exactly double speed, draining MixDesk X's buffers twice as fast as they filled.
• MixDesk X now opens such devices in their native 32-bit format and converts itself — the broken path is never in the signal chain. Float-native devices and Windows were never affected.

✦ How it works

• Where: open a bus EQ in 31-band mode → AUTO-EQ → pick your measurement mic channel → click 🎤 CAL and select the calibration file. The file name appears next to the button; × removes it.
• Supported format: plain-text frequency/dB pairs — exactly what miniDSP ships with the UMIK-1 and what REW uses (.txt, .cal, .frd). Header and comment lines (*, ", ;, #, "Sens Factor") are skipped automatically; comma decimals are fine.
• What it does: the curve is interpolated (log-frequency) onto the 31 ISO terz bands and subtracted from the analyzer readings of that channel — the live RTA display AND the Auto-EQ measurement both use the corrected values, the same way REW applies cal files.
• Scope: per measurement channel, active for the running session. No file loaded = uncorrected readings, exactly as before.

✦ Features

• 31-band terz EQ (ISO 1/3-octave, 20 Hz–20 kHz) as a fourth EQ mode on every channel and bus — switching 11↔31 band resamples the curve so the shape survives
• RTA: live 1/3-octave spectrum rendered directly behind the EQ bands — same ISO bands, aligned 1:1, zero CPU cost while the editor is closed
• Auto-EQ (bus editors): pink noise on the bus, measured through any mic input channel, averaged and inverted onto the 31 bands — boost capped at +6 dB so it never fights a room null, cut down to −12 dB, cancel restores everything
• Rename channels with a double-click on the name — name your measurement mic properly

⚠ Fixes

• New channels are no longer all named "CH 1" after loading a preset — numbering continues correctly

⬢ Highlights

• Cable-free system & per-app audio capture on both platforms. Windows 10 2004+ via WASAPI Process Loopback, macOS 13+ via ScreenCaptureKit — no virtual cable, no driver. MixDesk X excludes its own output, so capturing can never feed back.
• CLAP plugin host in every strip — native plugin GUIs, state saved in presets, one-click install of the free Airwindows Consolidated pack (~400 effects).
• Per-bus delay 0–1000 ms — sync Bluetooth and wired speakers playing the same source.
• Free. No account, no ads, no Pro tier. 9 UI languages. Signed & notarized.

★ New in v1.4

• iPhone as a wireless mic for LLT Desktop — stream your iPhone mic to the Mac over Wi-Fi and remote-start it in Translate or Dictate mode. A reliable replacement for Continuity.
• Dictation mode — spoken text is typed straight into the focused app on the Mac (via the companion), or saved on the iPhone as a note: original, translation, or both.
• iCloud sync — transcripts and dictation notes are stored in iCloud Drive (the “LLT Mobile” folder) so they appear on your Mac automatically.
• Built-in TXT/SRT viewer — tap a saved file to read it; multi-select to share or delete.

★ New in v1.3

• OBS Browser Source overlay — host live subtitles on your local Wi-Fi for OBS on your PC. Add a Browser source in OBS pointing to the URL shown in the app. Works with any browser-source-capable software (Streamlabs, XSplit, vMix).
• VRChat output (OSC) — send subtitles straight to the in-game chatbox over OSC, like Kikitan. Pick original, translation, or both. Test button + optional notification sound.
• Fully customisable overlay look — background and text colour, opacity, and border, applied live in the app and in the OBS browser-source overlay.
• Toggle the original text on or off — show it alongside the translation, or give the translation the full space.

★ Stream Deck + / + XL (Companion export)

• New export layouts: Stream Deck + (8 keys, 4 dials, touchstrip) and the new Stream Deck + XL (36 keys, 6 dials, touchstrip) — the dials and the strip are actually used, not just the keys.
• Audio controls on every page: your volume groups from the plan plus optional Master, Bus A and Bus B land on the rotary knobs — on every exported page, because the knobs are always there. Turn = volume ± step (2/3/5/10 % selectable), press = audio toggle (mute). The touchstrip carries the labels; with more controls than knobs the assignment cycles from page to page. Decks without dials get dedicated CTRL pages with Mute / − / + buttons instead.
• OVL page: the overlay channels assigned in the plan become toggle buttons plus an ALL OFF button.
• Why generic-http and not the Companion vMix module? A deliberate choice: one generic-http instance with plain vMix API calls works in every Companion version, today and years from now. vMix's relative volume API (SetVolume +=/−=) means the knobs need no state at all — and the levels are visible where you're looking anyway: the group knobs in the timeline bar and the Master/Bus faders in the CTRL panel are synced with vMix in real time. The Stream Deck is for eyes-free operation; DP is the display.

The exported Stream Deck + XL layout in Companion: cue buttons with real input thumbnails, dial rows for vMix Call, Master and Bus A/B underneath.

★ vMix categories — switch them from anywhere

• New cue command “Switch Category”: a cue can switch the category tab in the vMix operator view — jump to “Callers” before the guest block. Choreography for the operator surface, not just the program.
• CAT switcher in the timeline bar: category buttons in their vMix colors, with your own labels (“Callers”, “Graphics”, …) saved in the plan. Custom categories appear once labeled.
• CAT page in the Web Controller and in the Companion export — flip the operator view from the couch or the desk.
• Honest footnote: the vMix API can only switch categories — it can neither read nor assign them (verified up to vMix 29). That's why DP keeps its own labels; the moment vMix exposes categories in the API, we'll read them in automatically.
• Fully localized in all 20 languages.

★ Thumbnails & connection

• Not connected? The fix is right there. The “vMix not connected” screen now offers on the spot: enter an IP and connect — or one click on “vMix runs on THIS machine”, which detects and fills in the machine's network IP automatically.
• Running DP on the vMix machine? Thumbnails are automatic. Built-in thumbnail server: previews appear with no install, no firewall prompt, not a single manual step.
• DP on a different machine? When previews are missing, the input list shows a hint with a direct download of the free DP Thumbnail Server, plus a step-by-step guide in Settings. Can be turned off permanently.
• Mock save asks before saving without preview images.
• Everything localized in all 19 languages.

★ Web Controller

• Generated from your plan. Like the Companion export, every vMix command becomes a touch button — timeline colours, optional thumbnail. But there is no file: the surface exists only while Directors Plan runs. Every tap runs the same command logic as a click in the app (Command-Log and Mock mode included).
• Set up in seconds. Settings → Web Controller: switch the server on (port configurable, default 8570; optional 4-digit PIN). The reachable address shows as a QR code — point the tablet at it. Enable once; the server remembers its state and starts with the app. The state is saved in the plan too (“on wins”; port and PIN always come from local settings).
• Always current, hands-free. On plan open the surface is generated and pushed to all connected devices; plan changes (new cues, overlay assignments) sync automatically every 30 seconds. The new 📡 button at the left of the timeline bar pushes instantly.
• Four pages. Cues (all commands as touch buttons), Replay (one row per time, Cam 1–4 / 8, a tap marks and plays the last N seconds to output), CTRL (volume groups as big touch faders), OVL (assigned overlay channels as toggles).
• Built for touch. Tap fires; hold (or right-click on desktop) opens Play/Pause & Restart for that input. Four button sizes (S/M/L/XL) that stay physically identical across devices — only the column count adapts. Each device paginates for itself so nothing scrolls; rotating re-paginates live. Wake-lock keeps the screen awake; a status dot shows the connection.

The Web Controller in a tablet browser: cue buttons with thumbnails in timeline colors, pages for Cues, Replay, CTRL and OVL along the bottom.

⬡ License & security

• The surface is served only by the running, licensed Directors Plan instance and is meant for the local production network; an optional PIN keeps curious fingers in the same Wi-Fi out.

★ New design

• One consistent interface across the whole app — a reworked, modern, minimal look.
• Dark and light mode fully aligned: timeline, cue editor, settings, exports and the Controller now all follow one shared colour system.
• Flatter, quieter styling: consistent buttons, input fields, dialogs and dropdowns.
• Focus rings only on keyboard use — no more distracting outlines when you click.
• Subtler scrollbars and a consistent typeface on every platform.

⬡ Right-click menus & dropdowns

• Context menus no longer clip at the bottom of the window — long menus scroll internally, and look identical in the timeline, transport bar and cue editor.
• On Windows, pop-up selects (e.g. the transition picker) now follow the chosen app theme instead of the system setting. On macOS and Linux these system menus still match the OS design.

⚠ Fixes

• Light mode: several wrong or hard-to-read colours corrected.
• Settings: the license overview shows the packages in multiple columns again on large screens.

★ Controller — a live control surface

• What it is. A new CTRL button in the timeline bar (same family as OVL/REP) opens a panel with five tabs — Groups · Audio Out · Stream · Output · Rec. Every button is an instantly-fired vMix action; with no vMix connection everything is cleanly disabled — no crash.
• Groups — the real gap. vMix has no group fader that rides several inputs at once (think multiple callers whose audio doesn't follow the video). Build free groups from any inputs — keyed on the GUID, so they stay stable — and one fader fires SetVolume on all members at once. The set level is saved in the plan. Round volume knobs per group sit right next to CTRL in the timeline bar for quick access.
• Sound on/off with fade. A per-group on/off toggle (mute to 0 / back to level) as a deterministic state toggle, with a free per-group fade time in ms (SetVolumeFade) — music bed 1500 ms, mic group 0 ms. Deliberately a level on/off, not vMix "mute", because the API mute is unreliable on call inputs.
• Audio Out & live values. Master and Bus A–G as faders that show the real vMix levels and move live — even when levels are changed elsewhere (in vMix, on a hardware audio desk, a Stream Deck). Powered by the vMix TCP API (pushed in real time) plus a one-shot actual-value read on open/refresh so the faders are right from the first second. Pin Master or any bus into the timeline bar for constant access.
• Stream / Rec / Output. Start/Stop per stream number (and "all"); Recording, External Output, MultiCorder and Fade-to-Black with live status dots and confirmation toasts. The Output matrix shows Fullscreen 1/2 and Output 1–4 with the currently active source (read from the vMix API XML), switchable for Fullscreen 1/2 and Output 2–4; NDI/OMT/SRT status shown as live LEDs.

⬡ Under the hood

• Status and outputs are read from the vMix API poll that already runs — no extra load. Real-time levels come over a separate lean TCP push channel (event subscription, not polling) plus a single actual-value read on start/refresh.
• Both Controller and Append are localized across all 19 program languages (German/English complete, the rest with English fallback).

★ DP-Append — add modules to a running show

• What it does. Merge several DP show modules into one combined show without replacing the show you currently have loaded. Mirrors the vMix bundle-append workflow.
• How you use it. A split Open ▾ button — primary click = Open (replaces the show as usual), the caret = Append. Keyboard shortcut ⇧⌘O (Open without Shift, Append with Shift). The file picker takes .dplan and .yaml and remembers the last folder used.
• How it works. The module's cues are appended to the end of the running show — the active show, its file path and the history stay untouched. Every appended cue gets a new, collision-free ID and is tagged internally as "appended" (source = module name). The anchor is always the vMix input GUID (InputKey) — never number or title — so cues stay correctly matched after an append.
• Live / Pending. Inputs that aren't in the vMix session yet are flagged PENDING — such a cue won't fire and shows a hint. Re-Scan inputs re-reads the vMix inputs and switches PENDING cues live.
• Confirmation. On success you get "Module "{name}": {count} cues appended".

★ New in this release

• Replay pages in Companion / Stream Deck export. The export dialog gets a new vMix Instant Replay section. Tick Include Replay pages in export and the generated Companion config carries dedicated replay pages with one button per cam × time preset (Cam 1 -5s / -10s / -20s, Cam 2 …). Each button fires the same three-call vMix sequence as the in-app REP panel — select cam → mark the last N seconds → play to output.
• Edition-aware grid. Choose 4 cams (4K edition) or 8 cams (Pro); the page fills only the rows it needs.
• Position in the page deck. Drop the replay pages in at the top (First page(s)) or after the vMix pages, so they fit whatever Stream Deck layout is already in use.
• One page series per preset. Optional toggle: generate one page per preset (Standard / Long / Quick) instead of just the currently-active one — the operator switches presets by flipping pages, no mode-toggle mid-show.
• Stop + page-nav inline. Column 0 of each replay page reserves Next / Prev between replay pages and a red STOP button to halt replay playback — no jumping back to a different controller page.
• Cam display names carried over. The cam labels from the DP app (e.g. "Robotic Wide", "Hand-Held 2", "Crane") render straight into the Stream Deck button captions, so the operator sees the same names live as in planning.

⬡ Under the hood

• Export schema extended for the Replay command type — opens the door for HTML and CSV export to follow.
• Companion config generation reuses the existing button-grid renderer; no second export path.

★ New in this release

• Instant Replay panel from the timeline. New REP button right next to the existing OVL button in the timeline toolbar. One click opens a collapsible panel with all replay cameras side by side — four on the 4K edition, eight on Pro. Each cam has three time buttons (-5s -10s -20s); a click fires the full vMix replay sequence in one shot: select cam → mark the last N seconds → send to program. Three sequential API calls, one click, no switching out of DP.
• Three time presets. The time buttons cycle through three predefined presets — Standard (5 / 10 / 20 s), Long (10 / 20 / 30 s), Quick (3 / 5 / 10 s). The active selection is remembered across sessions.
• Recording toggle in the panel header. A REC indicator that doubles as a switch: grey to start recording, red to stop (with a safety confirm on stop so a running recording can't be killed by accident). Recording state is read live from the vMix XML — while recording is off, the cam buttons are cleanly greyed out with a hint, instead of firing into the void.
• Keyboard for the fast hand. While the panel is open: , . / fire the three preset times of the most-recently used cam, Esc stops replay playback. Cam selection by number keys is deliberately left out — 1–9 in DP are already bound to preview switching.
• Edition-aware. The REP button only appears when the connected vMix instance can actually do Instant Replay (4K or Pro edition) and a Replay module is active. On HD, SD or Basic editions the timeline toolbar stays unchanged — no dead button, no confusion.
• Runs alongside auto-playback. The replay workflow does not block the live show. While Directors Plan auto-plays the rundown, you can drop in a replay at any time without interrupting cue playback — vMix transitions back to the program automatically after the replay, and the plan keeps running.

⬡ Under the hood

• vMix XML polling was extended with the replay block (lives inside the Replay input, not at the top level) — no second poller.
• Edition is now correctly read as an XML child element (it was misinterpreted as an attribute before — a latent bug).
• UI strings added for all 19 supported languages.

⚠ Bug fixes

• Auto-calc default on fresh installations. Timeline auto-calc was disabled by default on fresh installs due to a legacy config flag. It now works correctly out of the box.

★ New in this release

• Video List — full integration. vMix only exposes the duration of the currently active item in a Video List. Directors Plan now measures each item on demand (~200–300 ms per item) by briefly cueing it to Preview, then restores the original index. Durations persist in the plan and in mock files and feed the timeline, auto-playback waits, HTML/Companion exports, and Streamdeck button labels. Three measurement modes from the timeline right-click menu: single item, from-item-to-end, or full list — with a cancellable progress overlay and 5-second per-call timeout. Dedicated Commands Builder category with Select Video List, Video List Item (index + optional transition in one call), and Next / Previous Item. Drag & drop of single items or whole lists from vMix Controls. Right-click editing of index and transition on existing commands, with the cue duration recomputed immediately. ⏮ / ⏭ quick-action buttons on timeline thumbnails. All 19 UI languages updated.
• vMix 29 compatibility. Stinger 1–8 (was 1–4) everywhere — Commands Builder, transport buttons, Settings default, auto-playback, timeline drops, CueEditor, Companion and HTML export. 16 overlay channels (was 8) with scrollable timeline popover and CH 1–16 cue-overlay assignments. Both overlay XML formats parsed in parallel so older vMix versions keep working.
• CSV rundown export — for client sign-offs. Built for the review round with the customer: many clients cannot open HTML files (corporate security policies), while CSV opens and edits instantly in Google Sheets, Excel and Numbers.
  • Configurable columns — every column toggled by checkbox. Headers come out in the active app language (Pos, Length, Live/Graphic, Segment, Location, Technical Notes …) using the same i18n keys as the rest of the app.
    • Default ON: Pos / Length (combined into one cell — Pos on the first line, Length on the second; each also toggleable individually); Live/Graphic (segment header with the user-chosen column name, e.g. "Topic", "Segment", "Sport"); Content (cue description); Commands — all vMix / HTTP commands of the cue in readable form, each one on its own line inside the cell, line-break separators render as real blank lines (just like in the timeline). Examples: Cut: Camera 1, Item 3 (03_football.mp4) → Stinger3 · List 1, HTTP POST https://hooks.example.com/show-start.
    • Default OFF: Location, Technical Notes, Overlay channels — switchable on demand.
  • Optional plan header above the cue table, each as its own block separated by a blank line: Basic Data (show, prod-no, version, date, from/to, length, plus all custom fields from "Basic Data Extra"), Schedule (multi-line), Contacts (all dynamic contact blocks with field labels and values).
  • Excel- and Sheets-compatible — UTF-8 with BOM so Excel renders umlauts, accents and Asian scripts correctly; separator selectable (comma — default / international / Google Sheets, or semicolon — for Excel with German locale where the comma is the decimal separator); CRLF line endings; proper quoting for values containing the separator, quotes or line breaks.
• Generic HTTP Request command. New Special command to hit any HTTP endpoint — not just vMix. URL, method (GET / POST / PUT / PATCH / DELETE / HEAD), body with configurable Content-Type. Companion export carries it through as the matching HTTP action, runnable directly from a Streamdeck button. Useful for light desks with web APIs, info displays, Slack/Discord show-start webhooks, or third-party streaming gear.
• Start from this cue — manual auto-playback override. Every cue row gets a small ▶ play button. First click arms it (orange + pulse, 5-second confirmation toast); second click stops any running auto-playback and starts immediately from that cue. While the cue is on program, the button turns green. Two-click gate prevents accidental jumps during live shows; the drag handle next to it stays fully functional.
• Under the hood. 5-second HTTP timeout via a real AbortController (browsers ignore the timeout option.
• Backwards compatible. Plans from v1.42.140 and older open without migration, conversion, or user action — all new fields are optional. Existing commands (Cut, Fade, Stinger 1–4, Overlay 1–8, SelectPlayList …) run unchanged; new command types appear only when you actively use them.

✦ Core features

• Visual timeline with drag-and-drop rundown creation and frame-accurate playback
• Full vMix integration — inputs, transitions, overlays, audio control
• Live override with shortcuts and re-triggerable cues
• Mock Mode for offline planning without vMix connection
• Stream Deck / Companion 3 export with thumbnails
• HTML, PDF & CSV call sheet / rundown export
• Companion App export for mobile crew reference
• 20+ supported languages
• 30-day free trial — every feature unlocked, no credit card required
• Multi-license discounts (3 seats: 15 %, 5 seats: 20 %, 10+: contact)
• Companion: DP Thumbnail Server — open source, free on GitHub

⬡ Platforms

• macOS Universal (Apple Silicon & Intel)
• Windows 10/11 — Intel/AMD64 + ARM64 installers

⬡ Improvements

• Menu bar app: German → English. Menu labels, popover strings and tooltips translated from German to English for the international release.

ℹ️ MDE — Mix Desk EQ is retired — its successor is MixDesk X (free, macOS & Windows). MDE 1.3.042 is the final version, keeps working, and stays downloadable here.

✦ Features

• Multi-channel mixer with flexible input channels and up to 4 stereo output busses
• 3 audio source types per channel: microphone/hardware input, system audio, single-app audio (ScreenCaptureKit)
• Per-channel: Gain (-20 to +40 dB), Constant-Power Pan, EQ (2/3/11-band), Mute & Solo, vertical Fader with RMS meter
• Per-output-bus: own Fader, EQ, Output Delay (0–1000 ms for Bluetooth latency compensation), Mute, color-coded
• Internal format: 48 kHz / 32-bit float, automatic sample-rate conversion
• Low-latency direct path (RingBuffer → SourceNode) for 48 kHz stereo inputs
• Adjustable buffer: 128–2048 samples (from 2.67 ms latency)
• Auto device recovery on disconnect (per-bus, no cascade)
• 3 power modes: Always On, ECO (auto-suspend after 60 s), ECO Sleep
• JSON-based preset system (Save / Load / Auto-load on launch)
• Menu bar app with mini-mixer popover
• Free Audio Driver Installer (BlackHole-based, up to 8 virtual stereo devices) — also on GitHub
• 14-day free trial, 30-day offline grace period after activation

⬡ Platforms

• macOS 13+ · Apple Silicon build (universal Apple Silicon)
• macOS 13+ · Intel build (separate installer)

★ New local engine — Parakeet (now the default)

• Parakeet (NVIDIA TDT v3, MLX, Apple Silicon) — a second, very fast local speech engine, now the default. Noticeably faster recognition, the first sentence appears far sooner.
• Continuous streaming — text flows along live instead of only appearing at the end of a sentence.
• Dedicated "Parakeet" tab with timing sliders — tune "pause response time" and "max. sentence length without pause" to your speaking style: speed vs. clean sentences.
• Parakeet auto-detects the language; Whisper remains the multilingual, language-pinnable alternative — pick per task.

★ Dictation polish

• Optional dictation polish — a local LLM cleans up punctuation and grammar of the dictated text (same language, no translation). Cleaner text, fully offline.

⬡ Overlay & output

• Optional live preview — your spoken original runs live in the overlay; the translation fills in once the sentence completes. Fixed positions, no jumping.
• Consolidated "Output" tab — overlay colors, OBS, vMix and VRChat in one place.

⬡ Setup & convenience

• First-run onboarding — choose your audio source (Mac mic or iPhone over the network) plus a permissions checklist with live status.
• Uninstaller for clean removal, and "Reset to defaults" in settings.
• Smarter defaults: Parakeet as default, target language = system language, original shown in the overlay by default.
• The OBS overlay URL now shows the LAN IP directly — easier OBS / browser-source setup.

★ Dictation mode — speak straight into any app

• Type into any app (⌘⌥D) — speak and the recognized (optionally translated) text is typed straight into whatever app has focus: Mail, Word, Notes, browser, editor. Global hotkey or the menu-bar menu. To type into other apps, macOS asks once for the "Accessibility" permission.
• Choose content & mode — insert original only, translation only, or both; pick "Final" (completed sentences) or "Live" (types as you speak, self-correcting).
• LocalAgreement — a streaming algorithm that brings smooth, flicker-free live typing even to fully local, 100% offline Whisper.

★ iPhone as a wireless microphone

• With the free LLT Mobile app, your iPhone streams its mic to the Mac over Wi-Fi and can remote-start dictation or translation on the Mac — a reliable replacement for Continuity.

⬡ More

• Automatic in-app backend update — an outdated backend is detected and updated safely, with automatic rollback and no new model downloads.
• OBS / browser overlay & VRChat (OSC) — customizable overlay colors; long sentences are no longer truncated.
• Localized interface (DE / EN / ES / FR / IT).

★ Subtitles now in OBS and VRChat too

• OBS overlay (browser source) — LLT serves a transparent subtitle page; just add it in OBS via "Source → Browser" using the shown URL. Translations appear live in your stream, in the same style as the on-screen overlay.
• VRChat output (OSC) — subtitles are written directly into the VRChat chatbox (like Kikitan). Choose whether to send the original, the translation, or both — including a test button and an optional notification sound.

⬡ Overlay improvements

• Long translations are no longer truncated ("…"). Text now wraps and the overlay window resizes its height automatically.
• Overlay colors are now fully customizable — background, text color, colored border and opacity. New, more visible default: semi-transparent light green with black text and a green border.

⚠ Bug fix

• "Show Original Text" works correctly again and persists across restarts.

⬡ Overlay improvements

• Long translations are no longer truncated ("…"). Text now wraps and the overlay window resizes its height automatically — everything stays readable, even with fast speakers.
• Overlay colors are now fully customizable — background, text color, colored border and opacity (Settings → Display & Audio). New, more visible default: semi-transparent light green with black text and a green border.

⚠ Bug fix

• "Show Original Text" now works correctly — the setting is respected and persists across restarts.

⚠ Bug fixes

• Transcript writing on cloud engines. TXT and SRT transcripts are now saved correctly when using the direct cloud engines (Google REST, Google gRPC, Deepgram, AssemblyAI, Azure). Previously the transcript file was not being written for these paths.

★ New in this release

• Google gRPC Streaming STT — new true-streaming path through the Python backend. Flow: audio → backend → Google gRPC STT (real streaming, no VAD required) → backend returns recognized text (skipTranslation: true, Mistral is skipped) → Swift calls Google Translate REST directly using the same API key → original + translation are displayed. Requires the backend running, google-cloud-speech installed in the Python venv, a Service Account JSON, and a Google API key for Translate. Restart the backend after enabling.
• STT Engine matrix — seven engines, each pairing a speech-to-text source with a translation provider. Some run fully local, some call the cloud directly, some go through the backend:

• vMix Title Output — send translations directly to vMix as live title updates via the Web API. Includes Lower Third .xaml template
• Smart Pause Detection (VAD) — waits for natural pauses, produces cleaner sentences and dramatically better translation quality. VAD is set automatically per engine — streaming engines detect pauses server-side.
• Whisper Model Selection — 5 models (tiny / base / small / medium / large), download/switch/delete on demand
• Network Access & LLT-Mobile — free iOS companion app: wireless mic or standalone audio source via Wi-Fi

✦ Core features

• 3 audio inputs: microphone, system audio, single-app audio (ScreenCaptureKit)
• 101 target languages with auto source-language detection
• Live floating overlay, TXT transcript, SRT subtitle export
• Text-to-Speech output, Auto-Gain, Manual Gain
• Mistral 7B 4-bit MLX for local translation, on Apple Silicon (GPU + CPU)
• Menu bar app — backend auto-start, status icon, popover control panel
• 5 UI languages: DE, EN, ES, FR, IT
• 14-day free trial, 30-day offline grace period after activation

⬡ Platforms

• macOS 13+ · Apple Silicon only (M1/M2/M3/M4)
• 8 GB RAM minimum (base model), 16 GB recommended (medium/large)

✦ Changes

• Free for everyone — the 5-minute trial, the $3.99 purchase and the license activation are gone entirely
• All features included: broadcast-quality compressor, master volume, 5 built-in presets (Standard, Social Media, Video/Streaming, Night Mode, Gentle) and unlimited custom presets
• Set & forget as before: pick a default preset once and every video on every page just sounds right
• Still no account, no tracking — no data leaves your browser

⬡ Platforms

• Chrome Web Store — Chrome, Edge, Brave and other Chromium browsers
• Mozilla Add-ons — Firefox
• Android via Kiwi Browser — load the Chrome version

✦ Features

• Professional Web Audio API DynamicsCompressor: Threshold, Ratio, Knee, Attack, Release, Makeup Gain
• Master Volume (0–200 %) — balance audio between browser tabs or against other apps
• 5 built-in presets (Speech, Music, Video, Night Mode, Gentle) plus unlimited custom presets
• Per-tab on/off toggle, real-time gain reduction meter
• Set a global default preset — every video on every page just sounds right
• 5-minute free trial, then $3.99 one-time (purchase inside the extension)
• No account, no tracking, no data leaves your browser

⬡ Platforms

• Chrome Web Store — Chrome, Edge, Brave and other Chromium browsers
• Mozilla Add-ons — Firefox
• Android via Kiwi Browser — load the Chrome version

⚠ Limitations

• DRM-protected content (Widevine, e.g. Netflix, Disney+) bypasses browser extensions and cannot be processed