Mixed-language sessions
A set of failures in sessions that hear two languages is fixed:
- Subtitles that never finished, sitting on screen until the next sentence replaced them.
- Translations that arrived all at once after nothing appeared for a while.
- Speech that disappeared when it arrived on the source set to the other language, and sentences left untranslated because they were mistaken for the language you translate into.
- Invented sentences in the wrong language, sometimes repeated across several rows.
One deliberate change: a source set to a specific language is now recognised strictly in that language. Use automatic detection for a source that should hear more than one.
Smaller updates from here on
Speech models no longer travel with updates. They stay with your app data, and updates after this one download about 420 MB instead of about 1.7 GB.
Fixes
- Cloud translation could stop mid-session claiming the translation limit was reached, with most of the allowance left. The real cause was the app failing to renew its sign-in, which it now does by itself.
Plus a number of smaller fixes to recognition and diagnostics.
If something does not work as expected after updating, let us know in Discord.