Anysub 4.0.2

This release updates Node.js and other dependencies with known security problems. Nothing changes in how Anysub works.

Security

• The web image runs Node.js 24.22.2, with the Node.js security releases up to October 2026.
• Updated the web libraries SvelteKit, Zod and devalue to versions with security fixes.
• Updated the worker libraries Pillow, filelock, certifi and setuptools to versions with security fixes.

Anysub 4.0.1

Anysub 4.0.1 is the first published release of the Whishper rewrite. It replaces Whishper 3 completely, under a new name, and cannot read Whishper 3 data. 4.0.0 lists what Anysub does, and Coming from Whishper 3 explains how to move over. This release fixes what a review found before 4.0.0 went out.

Security

• The sign-in limit now holds when many attempts arrive at once.
• Asking for a password recovery code again prints the same code until it expires or is used, so nobody else can keep replacing it.
• A server error during sign-in is no longer reported as a wrong password.

Fixed

• Library and uploads
  • An upload that stopped shows as Upload paused instead of keeping the library refreshing for a week.
  • The new project dialog checks the file type and size before sending anything, and explains a failed upload in plain words.
  • Leaving or reloading the page during an upload asks first, and a stopped upload no longer pulls you to its project later.
  • A finished upload cannot be sent twice, and the dialog no longer closes on a repeated Escape or a text selection dragged out of it.
  • Renaming a project keeps your place on the page, and reopening the rename no longer shows the previous error.
  • Library search ignores accents, so cafe finds "Café".
  • Cancelled projects are labelled Cancelled, and long names are shortened without breaking an emoji.
• Studio
  • No edit can move a cue past the end of the source anymore. Such a cue made every save fail; one that already exists is flagged Ends after the source.
  • Jumping to a cue that starts where the previous one ends selects that cue, so I, O and other edits change the right one.
  • I, O and split use the exact playhead during playback.
  • Moving a cue shifts its word timings with it, and changing its start or end clears them, so the JSON export never has words outside their cue.
  • On a narrow screen, following playback scrolls the cue list instead of the page.
  • Screen readers no longer read out the playback clock, progress and save state as they change.
  • Timeline ruler labels show fractions of a second when zoomed in.
  • A refused split shows as a notice instead of a save error, and holding Shift+Delete deletes one cue.
  • While media is prepared or transcribed, the studio checks progress with a small request instead of reloading the whole transcript every 1.5 seconds.
• Models and workers
  • Transcribing with a model that is not downloaded yet fetches it the same way the models page does, with a free-space check, progress and a working cancel.
  • A model download that fails resumes on retry instead of starting over, and a cancelled one frees its space.
  • Changing the language between runs no longer reloads the model, and an unsupported language fails with a clear message.
  • The models page shows a finished download or removal at once, and judges fit at the configured precision.
  • Removing a disconnected worker also cancels its running model tasks, and failure messages fit downloads and removals.
  • Media preparation interrupted by a crash or a short web restart completes on retry instead of failing or starting over.
  • Thumbnails of small videos are no longer upscaled.
  • The worker stops promptly when its container stops during a transcription, reports a job it cannot read instead of letting it expire, and checks its name and identity at startup.
• Running Anysub
  • An empty DATA_ROOT counts as unset.
  • Releasing a fix for an older version line never moves latest or the major image tag backwards.

v3.1.4

• Update yt-dlp

v3.1.3

What's Changed

• Fixed issue with VTT format (#81)
• Update README.md by @miketwo in #75

New Contributors

• @miketwo made their first contribution in #75

Full Changelog: v3.1.2...v3.1.3

v3.1.2

What's Changed

• Add CPU_THREADS environment to configure the CPU threads used. It defaults to 4. (#28)

Full Changelog: v3.1.1...v3.1.2

v3.1.1

What's Changed

• Use actual filename of uploaded media instead of "subtitles.srt" when downloading transcription by @JonasCz in #69
• Display time in hh:mm:ss format instead of seconds by @JonasCz in #70
• reversed transcription list display by @JonasCz in #72

New Contributors

• @JonasCz made their first contribution in #69

Full Changelog: v3.0.1...v3.1.1

v3.1.0 - large-v3 support

Added support for Whisper Large v3 model

• This release bumps the faster-whisper version to 0.10.0, and adds support for the new OpenAI large-v3 model (#48) .

More info about the model:
openai/whisper#1762

Full Changelog: v3.0.0...v3.1.0

v3 - 🐳 Container Consolidation

Warning

BREAKING CHANGES
You must update your docker-compose.yml file to the latest version available in this repository to ensure compatibility with the newest Whishper release (v3). For detailed guidance, refer to the updated documentation or opt for a fresh installation using the get-whishper.sh script.

GPU Version Users: Please follow these specific upgrade instructions to transition from v2.x to v3.x.

Key Updates in This Release

1. 🐳 Container Consolidation: I've streamlined the architecture by merging the three previous Whishper containers into a single container. Additionally, the whishper-proxy container is now redundant, as its functionality is integrated within the main container. As a result, the nginx.conf file is no longer required, simplifying the docker-compose.yml setup for a more user-friendly experience.

2. 🔄 File Upload Progress Bar: Addressing a common user concern (#20), I've added a progress bar for file uploads. This enhancement eliminates the uncertainty experienced during large file uploads where previously no feedback was provided, potentially giving the impression that Whishper had become unresponsive.

Upcoming Feature Teaser: In the next release, look forward to a highly requested feature – the ability to use local folders as media inputs (#15).

---

Changelog

• Simplified Docker Experience: Whishper has evolved into a unified docker container, requiring only a database container and the optional libretranslate for translations.
• Enhanced User Feedback: The new upload progress feature (#20) now provides real-time feedback during file uploads, a feature much awaited by our users.
• Update translation models: When you add new languages to the LT_LOAD_ONLY variable in the .env file, now LibreTranslate will update and download any new languages. This is achieved by adding LT_UPDATE_MODELS: True by default to the Libretranslate environment in the docker-compose.yml file.
• Additional Tweaks and Enhancements: This release also includes various minor fixes and improvements for an overall smoother experience.

Explore the Full Changelog: Whishper v3.0.0 Changes

v2.1.1-beta

fix: no space left while building

v2.1.0-beta

Merge branch 'main' of github.com:pluja/whishper