Add "listening" indicator when using free speech-to-text models

[ComputelessComputer]

because we're batch processing, it makes sense to show that sth is happening under the hood