Currently the Voice Recognition logic is all initialized at the start of the app, and this may work fine for you, but in some cases there may be missing voice packs for the configured locale settings on the operating system.
This was reported as part of #96 and could stop certain parts of flows not being registered correctly, while we cant entirely solve the problem of missing voice packs on the OS we can at least handle a failure gracefully and provide more logging around why it failed.
Given voice recognition is one of the main problem points for cross platform and is not often used, it may end up getting put into its own plugin and taken out of core going forward, this would allow windows users to get the plugin continuing to use the windows one, or maybe a VOSK plugin that would work for both, but that's tomorrows problem.