The tool is Ok for now although I am not really sure about credit consumption for every single action. That shouldn't be the case.

However, what would take this tool to another level would be if you had multi-lingual voices powered by Large Language Models (check naturalreaders-dot-com). I think that's what is lacking and would really make a difference.

For both the cases above, employing a BYOK strategy could minimise your cost and maximise user experience and result!