Skip to content

NowCaption 1.2.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 27 Oct 02:12
· 4 commits to main since this release
v1.2.0
e848db6

Live captions for any audio on your Linux desktop, recognised entirely on your own computer. This release brings NowCaption's new speech model, v3: fewer mistakes in everyday speech, and captions that keep up after pauses.

Install

With automatic updates (recommended):

flatpak install --user --from https://techlm77.github.io/NowCaption/nowcaption.flatpakref

Installed that way already? flatpak update gets 1.2.0.

Or download a file below — x86_64 for most PCs, aarch64 for 64-bit ARM — then:

flatpak install --user NowCaption-1.2.0-x86_64.flatpak

Or the AppImage, for Ubuntu 22.04, Debian 12 and newer — no installation needed:

chmod +x NowCaption-1.2.0-x86_64.AppImage
./NowCaption-1.2.0-x86_64.AppImage

Gear Lever or AppImageUpdate can keep it up to date.

What's new in 1.2.0

  • Speech model v3. NowCaption's own model, trained further so it keeps working after pauses, as the app hears speech:
    • About a fifth fewer wrong words when captioning continuously: 17.3% → 13.3% across meetings, calls, podcasts, accents and noisy recordings. Meetings: 25.8% → 16.4%.
    • The first words after a pause are wrong far less often: 28% → 17%.
    • Fewer mistakes for 16 of the 20 accent groups tested (all clips: 17.2% → 13.8%), and none clearly worse.
    • Words show a little sooner on average (13 ms at the median), with about the same CPU use.
  • Known limits:
    • A short sentence straight after a pause can show slightly later than before (about 26 ms at the median).
    • It needs about 30 MB more memory, and the download is 127 MB (was 101 MB).
    • Some targets set before training were not met: CPU (17.7% of one core, against a target of 15%; 1.1.x used 17.2%), word errors (13.3%, against 12.5%) and the delay target. The benchmark page lists every target, the full results with their ranges, and how they were measured.
  • Benchmarks you can check. The benchmark scripts are in benchmark/, and nowcaption-1.2.0-benchmark-results.tar.xz below has the per-clip results behind the tables: every clip's word errors and word times, and the recognised text where the test set's licence allows it. The reference transcripts are left out, as some test sets' licences don't allow passing them on.

Everything else works as in 1.1.16. The .zip and .tar.gz below contain the source code. The speech model is downloaded separately from the models-v3 release; its training data and licences are listed in MODEL-LICENSES.md.

SHA-256 of the Flatpaks, AppImages and benchmark results:

  • NowCaption-1.2.0-x86_64.flatpak: 26ca0c9bb0edb0fd53f128b0e19d289a3843ae67d71014b0767a62305d01e498
  • NowCaption-1.2.0-aarch64.flatpak: 9ebaabac878df22659afa2f51dc0a167dc41ad6d84482b9d2f0424ee0b095d51
  • NowCaption-1.2.0-x86_64.AppImage: 06ae508249ea0b20acdb3fe6f3951713a02cb3f8dc32ddc16564c7d60d802d60
  • NowCaption-1.2.0-aarch64.AppImage: 8d3db650a0196d95184e5a24099091639fe8f94b85dc1bc481ed115eb81ff9a9
  • nowcaption-1.2.0-benchmark-results.tar.xz: c09fb79dff4f8d691d256e7c8e581209e33c207b5d2ae18724fe74d7298f466b
Assets 9
Loading