Something that does work for me in this case is to redo the commands to 90% instead of '90 percent' or '90 per cent'. As Pfeil has stated, you've kind of run into an area of Windows speech recognition that can get a little squirrelly, as the speech engine doesn't have, 'to 90 percent' as a dictionary entry and is not getting things right. If you DO have '90%' as the entry, and the speech engine is getting, 'to 90 per cent', you may want to say, 'percent' a little faster, as it looks like it is picking it up as two words.
I got to sit in on the new HCS, 'Apollo' recording and got to try on the headphones - you can hear almost every single sound your face makes, so, it's possible your mic could be just a little too hot.