Speakr v0.10.14-alpha

In this release, the light or dark mode and the color scheme are saved with the account, every speaker can again be chosen for a segment in the transcript editor, and recordings can no longer be left stuck in QUEUED or SUMMARIZING.

Added

• Theme saved with the account (discussion #419). The light or dark mode and the color scheme were kept only in the browser, so they were lost in a private window or after clearing site data, and were not carried over to other devices. They are now also saved with the account when they are changed, in the main app, Inquire, the header of the other pages or a shared recording, and the saved setting is applied on every device. Each setting is saved separately: when the color scheme is changed on one device, the dark mode on other devices is not affected. For existing accounts, nothing is saved until the mode or the scheme is changed, and until then, the current setting of each browser is used. Thanks to @devkit415 for the suggestion.

Fixed

• Speaker list in the transcript editor (#425). The list was filtered by the name already in the field, so only the assigned speaker was offered, and since v0.10.10-alpha, choosing a name had no effect. All speakers in the recording are now listed when the field is opened, the list is filtered only while a name is typed, and the chosen name is assigned to the segment. Thanks to @vampire337 for the report.
• Recordings stuck in SUMMARIZING (#420). When no text model was configured, the recording stayed in SUMMARIZING after a summary was requested. The recording is now completed, with the same steps as after any other summary. Thanks to @jjsmackay.
• Recordings stuck in QUEUED (#423, #424). When reprocessing could not be queued, the transcript had already been removed and the recording stayed in QUEUED, where it could be neither reprocessed nor reset. The settings are now resolved before anything is removed, and a recording that cannot be queued is marked as failed and can be reprocessed. A recording in QUEUED can also be reset, and its waiting job is cancelled. Thanks to @jjsmackay.
• Errors from WhisperX (#422). Long error responses were cut at the end, where the exception is named, so only the first lines of the traceback were shown in a stored error. They are now shortened in the middle. Thanks to @jjsmackay.
• "Authentication Error" for other errors. A status code was recognised anywhere in an error, so a WhisperX out-of-memory error was shown as an authentication error because "401" occurred in an ID. A status code is now recognised only as a separate code, such as "status 401" or "HTTP 429".
• Conversion logs (#421). A file that became larger during conversion was logged with a negative "reduction". It is now logged whether the file is smaller or larger. Thanks to @jjsmackay.

Upgrading

Pull the new image and start as usual. Two empty columns are added to the user table at startup (theme_dark_mode, theme_color_scheme); no configuration changes are required.

WhisperX: in whisperx-asr-service v0.4.3, optional voice activity detection settings are added for text such as "Thank you." in silence or background noise (#418). They are set in the service's .env; no change in Speakr is needed.

Full Changelog: v0.10.13-alpha...v0.10.14-alpha