Skip to content

Hushcue 0.2.0

Choose a tag to compare

@github-actions github-actions released this 23 Oct 07:32

What's new in 0.2.0:

  • Subtitles read like professional ones: lines and subtitles break where the sentence does (after a comma, before
    "and"), not after "the" or between a first and last name, and each subtitle stays up long enough to read.
  • Subtitles no longer appear before anyone speaks: word timing is trimmed to the speech itself (in the test video,
    subtitles now start within 0.1 s of the speech 95% of the time, up from 2%).
  • Full stops and capitals no longer go missing in long stretches of speech, and fewer words are misheard (in the
    test set, 356 of 372 tricky words right, up from 348).
  • "Thank you." and "you" no longer appear over noise, tones or silence; "Thanks for watching" is kept when someone
    really says it.
  • The window fits any screen: it opens at the right size for your display scaling, can be made much smaller
    (down to 600×500), and every page rearranges itself for a narrow or short window instead of cutting things off.
  • On a narrow window the options move under the file list, the review editor puts the subtitles under the video,
    and buttons wrap onto a second line; a window too short for a page scrolls it.
  • The subtitle shown over the video in the review grows and shrinks with the picture.
  • Fixed: long file names and counts running under other text, the timeline drawing past its edges, the options
    card stopping short of the right edge, the setup screen's download text overlapping, and the setup card sitting
    at the top of the window instead of in the middle.
  • For developers: the app's CLI, tests and engine build on a Mac (CPU only), Windows and macOS builds of the
    engine use separate folders so one checkout can be shared, and a speaking-styles test video is made with macOS
    voices (tools/bench/make_speaking_styles.py); Hushcue.Cli timing scores word timing against it, and
    docs/subtitle-quality-research.md has the research behind this release.

Hushcue 0.2.0 for Windows 10 (1809) and 11, x64: subtitles from any video, made on your own PC.

  • Hushcue-0.2.0-setup.exe: installer. Installs for the current user (no admin needed) and can add
    "Generate subtitles with Hushcue" to the right-click menu of video and audio files.
  • Hushcue-0.2.0-win-x64.zip: portable copy; unzip anywhere and run Hushcue.exe.

Hushcue needs ffmpeg on PATH (winget install Gyan.FFmpeg). On first run it downloads the speech model
(about 1.6 GB) and, with an NVIDIA GPU, NVIDIA's cuBLAS (about 630 MB); without a GPU it runs on the CPU.
Installed copies update themselves from here; check the files against SHA256SUMS.txt.
The installer is not code-signed yet, so Windows SmartScreen may warn: choose More info, then Run anyway.