Repository navigation
Releases
Releases · eneskaramanoglu/lingocue
LingoCue v1.1.0
Added
- Template gallery (Stil). Every template is a card with a thumbnail of how it looks on your video: the
selected line over the frame where it is spoken. Thumbnails are drawn by the same subtitle renderer as the
preview and the export. A click applies the template; Ctrl+Z undoes it. The card that matches the project is
marked. - Rounded block box (Stil → Arka plan kutusu → Köşe yuvarlaklığı).
- The box is sized from the text measured in the font file. On test lines it lands within a pixel of the
square box. - Lines that cannot be measured (missing font) or are too long for the margins keep the square box.
- The box is sized from the text measured in the font file. On test lines it lands within a pixel of the
- Doubtful words. On a recognized track, a line waiting for review shows where speech recognition was
unsure: warning-colored bars on the timeline at those words, and buttons under the editor that play from
just before each word. - Apple Neural Engine (Tercihler → Genel, Apple Silicon). Whisper's encoder can run on the Neural Engine.
The Core ML encoder (~1.2 GB) downloads once; the first run can take a few minutes while macOS prepares it. - Prompt cache. After a correction or translation the status says how many tokens Gemini served from its
cache and about how much that saved. Gemini only caches when the shared start of the prompt is long enough
(large glossaries and contexts). - Website (
website/, Turkish and English) for Vercel: features, screenshots, privacy and costs, downloads and FAQ. - Releases also carry
LingoCue-osx-arm64.dmgunder a fixed name, so the latest Mac download has a stable link.
Fixed
- The export dialog showed "large-v3-q50" for a track named "large-v3-q5_0" (the underscore was read as a
keyboard shortcut marker).
Changed
- The main window's view model is split into smaller parts; the planning logic of runs moved into tested library
code. Nothing changes in use.
LingoCue v1.0.0
The first version packaged for everyday use: installers, updates and help.
Added
- Windows installer.
LingoCue-win-Setup.exeinstalls per user in%LocalAppData%\LingoCuewith
no admin rights, adds Desktop and Start menu shortcuts, and uninstalls from Windows Settings. Your
settings and models are kept.LingoCue-win-Portable.zipruns without installing. - Automatic updates.
- At start-up, at most once a day, the app checks GitHub for a new version. Only the version number
is asked. You can turn this off in Tercihler → Genel. - An installed Windows copy downloads the update in the background (usually a small delta) and
offers "Yeniden başlat ve güncelle". It asks about unsaved work and running exports first. - On macOS and with the portable zip, a banner links to the release page.
- Yardım → Güncellemeleri denetle checks right away.
- At start-up, at most once a day, the app checks GitHub for a new version. Only the version number
- macOS disk image.
LingoCue-x.y.z-osx-arm64.dmgwith a link to Applications, and an app icon. - App icon on both systems.
- Diagnostics package (Yardım → Tanılama paketi oluştur): a zip on the desktop with a summary,
recent logs and settings, ready to attach to a bug report. The home folder path is hidden, the
recent-files list is left out and API keys are never in it. Nothing is sent automatically. - Unexpected errors are shown with an offer to make the package; the app keeps running so you can
save your work. - Help menu: user guide, log folder, bug report on GitHub, and "LingoCue hakkında" with the version
and the licenses of the components. - User guide in Turkish: docs/KULLANIM.md.
Changed
- Release downloads are now an installer (Windows) and a disk image (macOS) instead of zip files.
Copies of 0.12 or older that ran from a zip do not update themselves: install 1.0 once with
Setup.exe or the .dmg.
LingoCue v0.12.0
Added
- Background box (Stil → Arka plan kutusu).
- Four kinds: none, block box (one rectangle behind all lines; the text keeps its outline), line
box (a box behind each line) and full-width band. - Box color with opacity, horizontal and vertical padding, and a frame with its own width and
color. - The preview and the burned-in video draw it the same way (libass). Exported as an ASS file, it
may look different in players that do not use libass; the panel says so. - The "Belgesel" template now uses the block box. New template: "Haber şeridi".
- Four kinds: none, block box (one rectangle behind all lines; the text keeps its outline), line
- Hardware detection. The welcome screen and Tercihler → Genel show the graphics card and its
memory (on a Mac: the chip and its memory).- "Önerilen ayarları uygula" picks the Whisper model, the accuracy mode and the local model batch
settings for this machine, and names the local model size that fits (4B to 30B).
- "Önerilen ayarları uygula" picks the Whisper model, the accuracy mode and the local model batch
- large-v3 (tam): the full Whisper model (3.1 GB) for GPUs with 6 GB or more.
- Whisper accuracy (Tercihler → Genel): "En iyi" uses beam search with five candidates. It is
more accurate (for example "Boğaz'ın" with its apostrophe) and about twice as slow. "Otomatik"
turns it on when Whisper runs on a GPU. - Glossary hints. Names in the project glossary go to Whisper so it spells them as written.
- Doubtful lines. Lines where Whisper was unsure, or that look invented (credits such as
"Altyazı M.K.", looping sentences, text in silent stretches), get the review mark (!). - AMD and Intel hardware encoders for export on Windows (AMF and Quick Sync), offered when they
work on the machine. - Local engine: lines per request and requests at a time are now settings.
Changed
- Whisper decodes only the speech it finds and skips long silences and music-only stretches. Times
are mapped back to the video. On a test recording this was about 30% faster. - On a Mac, Whisper runs on the GPU through Metal. The device shown is the one whisper.cpp
actually used (for example "Apple Metal (GPU)", "Vulkan (GPU)" or "CPU").
Fixed
- A sentence that came after a long silence could be missing from the transcript.
- Importing an ASS file with BorderStyle 4 (the libass box) lost the box.
- An apostrophe or comma no longer makes a correct word look doubtful.
LingoCue v0.11.0
Added
- Korean. Translations can now go into Korean.
- Korean tracks follow the Netflix Korean style guide: 16 characters per line, at most 12 per
second, with Latin letters, digits, spaces and punctuation counting half. - The prompt asks for the polite speech level (해요체) unless the context calls for another,
names written in Hangul, and standard Korean spacing.
- Korean tracks follow the Netflix Korean style guide: 16 characters per line, at most 12 per
- Several target languages at once. In Çevir, pick target languages as chips, for example
English and Korean. Each language becomes its own track; the video context is read only once.
Japanese is new in the list too. - Korean fonts.
- The style panel warns when a Korean track's font has no Hangul and suggests one that is
installed (Noto Sans KR, Malgun Gothic, Apple SD Gothic Neo, …). - In bilingual mode the second track has its own font. A Korean second track gets a Hangul font
automatically, so the preview and the burned-in video do not show boxes.
- The style panel warns when a Korean track's font has no Hangul and suggests one that is
- Local engine (Ollama). Correction and translation can run on a local model: free, offline and
without rate limits.- Tercihler has the default engine, the Ollama address and the list of installed models.
- Düzelt and Çevir can switch engine for each run.
- The welcome screen shows whether Ollama is running.
- Free-tier mode (Tercihler): one request at a time, batches of at most 15 lines, and a pace
that respects the per-minute limit. - Altyazı → Başarısız satırları yeniden dene sends again only the lines the model did not
finish, so lines that already finished are not paid for twice.
Fixed
- A correction or translation no longer stops and loses its finished lines when the model stops
answering.- A batch with no answer is split in half and retried. A line that still fails is marked and the
run goes on. - After three failures in a row the remaining lines are marked without calling the model again.
- Rate limits are waited out for as long as the API asks.
- A batch with no answer is split in half and retried. A line that still fails is marked and the
LingoCue v0.10.0
Changed
- New look ("Stüdyo"): a dark-first design with a matching light theme. It uses one accent
color, cards with rounded corners, quieter buttons and sunken input fields. - The menu and toolbar become one app bar:
- The document name has a dot when there are unsaved changes.
- Transkribe → Düzelt → Çevir form one workflow group; a finished step shows a green check.
- Undo, redo and settings are icon buttons; Dışa aktar is the primary button.
- The theme moved to Ayarlar → Tema.
- Welcome screen when no project is open: a drop zone with Video aç / Proje aç, the recent
files as cards, and a readiness list (FFmpeg, Gemini key, Whisper model). - The lines are a list instead of a table. Each row shows the time range and duration, the text,
and the reading speed as a green, amber or red label; the selected line has an accent edge. - The right panel has segmented tabs, and Kontrol shows how many problems there are.
- Kontrol shows error, warning and clean-line counts as tiles, and each issue as a row with a
severity badge. - The timeline follows the theme. The selected line is drawn in the accent color, and warnings
and errors are tinted. - The transport bar is its own card with a round play button. A label over the video shows the
resolution and frame rate. - Dialogs have an icon header, a footer bar with Vazgeç before the main action, and boxes for
cost estimates. Dışa Aktar picks the export kind from three cards. - The status bar shows a running job as a thin progress bar in a pill.
LingoCue v0.9.0
Added
- Undo and redo (Düzen menu, Ctrl/⌘+Z, Ctrl+Y or Ctrl/⌘+Shift+Z) for every project change:
text, times, adding/deleting/splitting/merging lines, style, tracks, quick fixes. Up to 100 steps.- Typing in one line becomes one step, and a timeline drag is one step.
- Saving remembers the state, so undoing back to it clears the unsaved mark.
- Timeline under the video: the waveform, the lines of the active track and the playhead.
- Click to seek. Drag a block to move a line, or drag an edge to change its start or end. Lines
never overlap their neighbours. - Ctrl/⌘+wheel zooms around the pointer and the wheel scrolls. The timeline follows playback.
- The waveform is prepared in the background and cached, so reopening a video shows it at once.
- Click to seek. Drag a block to move a line, or drag an edge to change its start or end. Lines
- Line shortcuts:
- Alt+←/→ moves the selected line by 100 ms; with Shift only its start moves.
[and]set the start and end to the playhead (AltGr+8/9 on a Turkish keyboard).- Ctrl+B splits a line. While typing it cuts at the caret; otherwise it cuts at the word spoken at
the playhead, or at the middle. - Ctrl+M merges a line with the next one, and Ctrl+Enter adds a new line.
- Yardım → Klavye kısayolları lists all shortcuts.
- Kontrol tab: every problem of the track (reading speed, line length and count, too short or
long, overlaps, gaps too small to notice). Picking one jumps to the line.- Quick fixes: Çakışmaları gider and Okuma hızı için süreyi uzat. Both can be undone.
- Bul ve değiştir (Ctrl/⌘+F): search with Turkish case rules (İ/ı) and an optional whole-word
match. Enter finds the next match and Shift+Enter the previous one. Replace one or all; replacing
all is one undo step. - Dosya → Son açılanlar: the last 10 projects and videos.
Changed
- Moving a line keeps at least 80 ms (2 frames) to its neighbour, the same gap the QA check and
automatic segmentation use, so editing does not create "gap too small" warnings. - Çakışmaları gider only changes lines that really overlap; lines that are merely close keep
their reading time and stay listed as a warning.
Fixed
- Saving preferences or a style template could freeze the app, because the save waited on the
UI thread. Files are now written directly. - Switching to the Stil or Kontrol tab no longer clears the selected line.
LingoCue v0.8.0
Added
- Çevir (toolbar, Altyazı menu): translates a track into a new track with exactly the same
timing.- The source defaults to the latest "Düzeltilmiş" track, because the edited text is
authoritative; what speech recognition heard goes along as a hint for each line. - Target languages: English (default), Turkish, German, French, Spanish, Italian, Arabic,
Russian and Azerbaijani. Translating a foreign-language video into Turkish works the same way. - Glossary entries for the target language are enforced.
- Turkish pronouns have no gender, so he/she/they is inferred from the video context and kept
consistent, or neutral wording is used. - The dialog shows the model and a token/cost estimate; an optional review pass gives more
natural wording. - The first translation becomes the bilingual mode's second track.
- The source defaults to the latest "Düzeltilmiş" track, because the edited text is
- İzler window (Altyazı menu, or the button next to the track selector, which is now always
shown): rename, change the language, reorder (export order), duplicate, add an empty track,
delete (with confirmation), and Zamanlamayı kaynaktan al.- "Zamanlamayı kaynaktan al" copies timing fixes made in the Turkish track into its translation.
- Deleting a track keeps the tracks made from it linked to the original lines.
- Importing a subtitle into a project that already has lines asks: add it as a new track (the
language comes from names likefilm.en.srt) or replace the active track.
Changed
- Reading speed limits depend on the track language: English allows 3 characters per second more
than Turkish (Netflix guidance: 20 vs 17 CPS). The editor's marks and the shortening pass follow it. - In a translation track the editor no longer marks every line as "changed"; the filter becomes
"Kontrol bekleyenler".
LingoCue v0.7.0
Added
- Dışa Aktar (toolbar, Dosya menu, Ctrl/⌘+E): one dialog for every export.
- Subtitle files: SRT (with a "Premiere Pro uyumlu" UTF-8 BOM option, on by default, and
import notes), WebVTT, or ASS with styling, karaoke and bilingual layout. Several tracks are
written as separate files. - Videoya göm (MP4): burns the subtitles in exactly as the preview shows them, using the same
ASS and the same libass renderer, with the project's fonts. H.264 with x264 "veryfast" at a
quality from "Küçük dosya" to "Çok yüksek kalite" (default CRF 19). The audio is copied, or
converted to AAC when MP4 cannot hold it. Variable frame rate phone videos keep every frame's
timing. - Fast hardware encoding: NVIDIA NVENC on Windows and VideoToolbox on macOS when they work on
the machine. A one-frame test encode decides, and the dialog explains why one cannot be used
(e.g. "NVIDIA sürücüsü eski: en az 610 gerekli"). - Açılıp kapanabilen altyazılı video: MP4 (mov_text, plays everywhere) or MKV (ASS tracks
with styling and attached project fonts). The video is not re-encoded; each track carries its
language and title, and the first one is the default. - Warnings before rendering: a missing font, a font without Turkish letters, an HDR/10-bit
source.
- Subtitle files: SRT (with a "Premiere Pro uyumlu" UTF-8 BOM option, on by default, and
- Export queue: video exports run one after another in the background. "İşler" in the status
bar shows progress, speed and time left, with cancel per job and "Göster" to open the finished
file in Explorer/Finder. A failed or cancelled export leaves no partial file. Closing the app
while exports run asks first.
Changed
- "Altyazı Dışa Aktar" (SRT/VTT/ASS submenu) is replaced by the export dialog.
- macOS needs
brew install ffmpeg-fullinstead ofbrew install ffmpeg: Homebrew's plain
ffmpegno longer includes libass, which burning in subtitles needs. LingoCue prefers
ffmpeg-fullwhen both are installed. With an FFmpeg without libass, the export dialog disables
"Videoya göm" and explains what to install.
LingoCue v0.6.0
Added
- Stil panel (toolbar button or the right panel's Stil tab): font, size, bold/italic/underline,
letter spacing, upper case, text/outline/shadow colors with transparency, outline or background
box, 9-point alignment and margins, fade in/out. Every change shows at once in the video preview,
which is the same libass output that ASS export and (in M6) burn-in use. - Templates: YouTube klasik, Netflix tarzı, Shorts / Reels (9:16) and Belgesel, using fonts
present on both Windows and macOS. Save the current look as your own template (shared by all
projects), import/export templates as JSON, delete your own. - Karaoke (word-by-word highlight) from the speech-recognition word timings: "Dolan renk" (words
fill with the highlight color as they are spoken) or "Aktif kelime" (only the spoken word is
highlighted, Shorts style). Corrected text is matched back to the recognized words, so karaoke
also works on the "Düzeltilmiş" track; tracks without word timings stay plain. - Bilingual mode: a second track (e.g. an English translation or an imported SRT) under the
active one, with its own size and color; the two never overlap (also with top alignment). - Fonts: the font list shows names exactly as the subtitle renderer finds them (e.g. "Arial
Black"), warns when a font is missing or lacks Turkish letters (ğ ş ı İ …), and "Font dosyası ekle"
copies a TTF/OTF into afontsfolder next to the project, which the preview uses (so a project
moved to another computer keeps its look). - Optional safe-area guides (93 % / 90 %) over the video.
Fixed
- Display-only text in lists could write itself back into the project (e.g. a track's language)
and, with formatted text, crash the app; such bindings are one-way now.
LingoCue v0.5.0
Added
- Düzelt (toolbar, Altyazı menu): an AI pass over the active track that fixes recognition
errors, proper names, punctuation and spelling by Turkish (TDK) rules. Timing is never touched;
the result is a new Düzeltilmiş track and the source track stays as it was. Lines are sent in
batches of 30 with neighbouring lines as context, answers are checked (JSON schema, ids, line
count, locked glossary terms) and retried with the problems listed; lines that still fail keep the
original text and are flagged. Optional second "reflect" pass and shortening of lines above the
reading-speed limit. Progress and İptal in the status bar; the result reports changed,
needs-review and failed lines, tokens and the estimated cost. - Before starting, the dialog shows the model, a token and cost estimate, and whether the model
has a free tier. - Context pass: one call summarises the video (topic, tone, sen/siz, speakers) and suggests proper
names and terms for the glossary. - Altyazı → Sözlük ve Bağlam: per-project glossary (term, preferred spelling, kind, note, locked)
and context, editable; suggestions are marked until accepted. - Ayarlar → Tercihler: default transcription engine, Whisper model and language; LLM fast and
quality models (automatic = newest stable Flash / newest Pro not older than Flash, picked from the
provider's live model list), batch size, parallel requests, reflect, shortening, circumflex
spelling; reading rules; API keys. Settings are stored insettings.jsonin the app-data folder
(no secrets). A note explains that Google may use free-tier content to improve its products. - Editor: a mark column shows changed (●), needs-review (!) and failed (✕) lines of a corrected track;
the original text is shown in the tooltip and under the editor, with Orijinale dön and
Kontrol edildi; Değişenler lists only those lines. Editing a flagged line clears its flag. - Gemini provider over
Google.GenAI(IChatClient), with a price table and transient-error backoff;
translation pipeline infrastructure (UI comes in M7).
Changed
- Whisper is the default transcription engine again (no API cost while developing and testing);
Gemini 3.5 Transcribe stays selectable in the dialog. - Ayarlar → API Anahtarları moved into Tercihler as a tab.
- CI and release use the
macos-15runner image; a release is published even when one platform's
package is missing, so it can be added later.
Fixed
- GitHub Release notes are taken from this changelog again (they only said "See CHANGELOG.md").