Releases: fmadore/Live-translation
Release list
v1.2.3
Save transcripts to your chosen folder with the native Windows Save As dialog. The last
successful folder is remembered. Export plain text, Markdown, SubRip (SRT), or WebVTT (VTT).
Cancelling a save keeps the transcript available and cancelling from the quit prompt keeps
the app open. Captions received during a save remain marked as unsaved.
For system audio, choose an output device or one open application. Application capture
includes its child processes and notifications; browsers may include multiple tabs.
Refresh the application list before selecting a source. If the application closes, its
capture stops without switching to all system audio. Select it again after restarting it.
The start and rehearsal buttons now share an aligned row, with readable explanations below.
SRT/VTT timing follows the caption session clock, including provider latency, rather than
alignment to a recording. Older recovered transcripts without timing can still be saved
as text or Markdown. Application selection must be repeated after relaunching the app.
v1.2.2
This release brings a bilingual interface, caption appearance controls, and more reliable audio capture and transcript recovery.
-
Use the interface in English or French, independently of the caption language.
-
Adjust caption typeface, size, colours, backing, and line width, with a contrast readout and an overlay placement preview.
-
Refresh audio devices without restarting and choose the output used for system-audio capture. Device changes update the lists automatically; disconnected sources offer an explicit retry or default-device fallback.
-
Read earlier transcript lines without incoming captions pulling you to the bottom. Jump to latest resumes following, and long transcripts use shorter paragraphs on screen and in exports.
-
Fatal provider failures release audio capture and finalize the last caption while a healthy second source can continue.
-
Audio tests no longer accumulate unused buffers, and recovery snapshots are saved atomically with ordered writes and deletion.
-
Overlay controls report failed appearance updates and preserve the confirmed placement mode.
-
Refreshing audio devices no longer fails when a worker thread already uses another COM apartment mode.
-
A recoverable microphone buffer underrun or overrun no longer ends an audio test or live session. Disconnections and invalid streams still stop the affected source.
The two reported Windows audio errors were retested successfully with the installed 1.2.2.0 ARM64 MSIX. Automated verification: 263 frontend tests and 63 Rust tests passed; the billable live-provider test remains opt-in.
For Microsoft Store submission, use Live.Translation_1.2.2.msixbundle, which contains both x64 and ARM64 packages. The individual .msix files are unsigned packages for local signing and sideload testing.
v1.1.0
chore(release): 1.1.0 Second live-subtitles engine (Gemini 3.5 Transcribe Live), a tray that keeps a live session reachable and makes Quit deliberate, and the transcript safety work. The three version files must agree and must exceed the version already in the Store, which Partner Center enforces on upload. 1.0.5 is published, so this is the first bump since the Store went live. The same bump was made on the throwaway test/rc-build branch to tell a sideloaded test install apart from the published 1.0.5 in Apps & Features; that branch is not merged, and this is the real one.
v1.0.5
v1.0.5 — accepted by the Microsoft Store The source that passed certification and is published at https://apps.microsoft.com/detail/9PFB8LR3RR9X (product 9PFB8LR3RR9X), native x64 and ARM64. 1.0.5 replaced on-device recognition with a deterministic bundled demonstration. The 1.0.3 submission had failed policy 10.1.2.10 — Start Subtitles did nothing on the review device — and none of whisper.cpp, Windows.Media.SpeechRecognition or the experimental Windows AI Speech/ML component worked on a machine nobody had configured. Dropping whisper.cpp also removed the C++ toolchain and made native ARM64 packages possible. Tagged at the merge of #34 rather than at main, so this tag is exactly the certified build: main has since taken wasapi 0.24 and cpal 0.18.2, which the Store's 1.0.5 does not contain.
v1.0.0
v0.6.0
v0.6.0 — operator and overlay redesign, move-mode keyboard controls, …
v0.5.3
v0.5.3 - fix Gemini live translation session setup
v0.5.2
v0.5.1
v0.5.1 — captions read as one line A sentence spanning two turns was drawn as two pills at two different sizes with a gap between them. The previous turn is now the dimmed lead-in of the same block as the live text, at the same size, so it reads as running text. Both share one character budget, so the caption block never outgrows a single caption's worth. In "Both" mode the lead-in is tracked per source, so one speaker's text can no longer attach to the other's caption.
v0.5.0
v0.5.0 — overlay always on top, cleaner transcripts - Overlay is re-raised above other topmost windows (Zoom, Teams, PowerPoint slideshow, browser presentation mode) every 500 ms, and never steals focus. - Saved transcripts are chronological paragraphs labelled "Microphone" or "System" instead of timestamped lines; the live monitor matches the export. - "System (Zoom)" is now "System audio" — loopback capture always covered every application's output.