Some things I noticed:
The transcription shows in real time, and shows in one of 3 colors. This must have been very tricky to implement and is implemented well, however, I find it distracting and it breaks me out of a natural flow of speaking.
Similarly the recording widget itself at the bottom is overly complex. Does it even need to be there at all? Can’t it be one/two buttons, and more embedded into the chat flow? When I start recording, it changes into a similar but also complex widget with two ways to delete all spoken so far and a back button that doesn’t work. There’s also a black widget that shows up when speaking english that I assume is following some unsaid convention aimed at aiding the user. Then there is karaoke mode?
Way too complex, you can whittle this all down to two buttons: record and help/hint.
Sometimes (often) I didn’t know what to say but a hint might have helped.
Finally there is no end state, do I just press back when all the objectives are done?
Edit: the topics of conversation can be quite… exotic
> Alex is visiting the zoo when he notices an animal missing from its enclosure. Sam, the zookeeper, needs help finding it.
Are the conversations themselves being generated by AI? I think the topics should be centered around things that happen in daily life.