Letterly

Overall it fits my needs. I could switch on multi speaker at the beginning of an offline recording and recorded a meeting of our old german scout guild for about 60 minutes. Different speakers were detected (3 times more than effectively there), and they sometimed spoke fragmented through each other, but when I stopped recording and uploaded it for transkription, I was impressed about what was recorded and transkripted. Then I used the meeting summary protocol feature. The result was ok, but lacking some final topic findings. Luckily there was an option for writing your own transformation prompt to advise the AI system on how to do. I wrote my own system prompt (in german) for the AI, tried and optimized it about 15 times. Now, the outcome is pretty amazing :-)
There are just 2 main points, which I miss so badly and could be improved:
1. Scrolling through the original transcription only via finger swipe up on such long transcription texts is boring. There should be implemented a better/alternative scroll methods.
2. Editing a transciption (deleting several characters und writing new text) does not stick the viewpoint on the typing, but scrolls some lines upward over an over again in most situations.
But overall it's a worthy tool.