Speakr v0.10.10-alpha

In this release, several recordings can be selected in the sidebar with the keyboard modifiers, and generation temperatures can be set by the admin. A case in which a recording could be lost during a previous upload is fixed, and pages no longer flicker on load.

Added

• Selecting several recordings in the sidebar. With Ctrl-click (Cmd-click on macOS), a recording is added to or removed from the selection. With Shift-click, every recording between the last one clicked and the current one is selected. In both cases, selection mode is turned on. The open recording is the starting point for Shift-click.
• Configurable generation temperatures (#411). The sampling temperature for summaries, titles, chat and event extraction can now be set in the admin dashboard, in the Default Prompts tab. Without a saved value, the earlier defaults are used: 0.5 for summaries, 0.7 for titles and chat, and 0.2 for event extraction. Output is more consistent at lower values, for example 0 for repeatable summaries with a local model. The temperatures for Inquire and speaker identification are unchanged. Thanks to @gguglietti for the suggestion.
• Unofficial companion apps. The README and the documentation now list two community apps: Minutes for Speakr (Android and Windows) and MeetingRecorder (Windows). Both are maintained by their authors and have not been tested by me.

Fixed

• Recording lost after an upload finished (#407). When a new recording was started while the previous one was still uploading, the crash-recovery copy of the new recording was cleared at the end of the upload, and the view was switched away from the recorder. Only the recovery copy that belongs to the uploaded recording is now cleared, and the view stays on the recorder while it is in use. Thanks to @bestonephoenix for the report.
• Flicker on page load. Pages were briefly shown unstyled and with untranslated labels. In dark mode, the main and Inquire pages were shown white before turning dark, and with a color scheme other than blue, a blue loading screen was shown briefly on the main page. The stylesheet is now built with the image instead of in the browser, and the translations for the user's language are included in the page. The light or dark theme is set before the page is shown, and the only loading screen has the page's own colors. In supported browsers, a short cross-fade is shown between pages. The Inquire page is also shown about 300 ms sooner.
• Header labels on the settings pages. On the Account, Admin and Group Management pages, translation keys could be shown in the header instead of labels until a menu was opened. The header is now shown once the translations are loaded.
• Inquire language. The Inquire page was shown in the browser's language instead of the language chosen in Account settings. The chosen language is now used, as on the other pages.

Other changes

• I added a browser smoke test. Every page is opened in headless Chromium and checked for a working header, translated labels, applied styles and script errors. The test is run on every change, and against the built image before the image is published.

Running from source

The Tailwind stylesheet is no longer loaded from a CDN. If you run Speakr from source instead of the Docker image, build it once with scripts/build_css.sh, and again after changing templates. See CONTRIBUTING.md.

Upgrading

Pull the new image and start as usual. No database or configuration changes are required.

Full Changelog: v0.10.9-alpha...v0.10.10-alpha