I'm still learning QML, and I am still trying to figure out what causes that - but I put that lower on priority because it doesn't affect the function.

I set it to be portrait-only for testing; I'm assuming you have a rotation tweak? I'll change it to allow landscape orientation so I can test that.

If it returns results near-instantaneously, then Pocketsphinx is doing the speech recognition. If not, then it gave up and used Google Voice for it - which means it had to upload a .flac file, the speed of which depends on your internet connection. (This can be disabled by editing /home/user/.saera if you prefer not to use Google Voice - I haven't made a QML version of the settings dialog yet.) There is, quite annoyingly, a bug in Pocketsphinx that can prevent it from recognizing the first utterance - which means that until Pocketsphinx had heard you once, it is going to be much harder for it to hear you. Now that I have the Harmattan port working, I'm focusing my attention on this bug, as it is the largest obstacle to usability.