It's actually not bad nowadays if you use APIs (e.g. Amazon Transcribe, Google speech to text, etc.). I worked with speech-to-text APIs for an AR app and they performed quite well. For a desktop app you could probably toss in something like Snowboy for custom hotword detection to make it function like Alexa.
I think it's mostly a matter of no one having bothered to write the software, because it's hard to think of an application where desktop voice control is super useful. Perhaps the first question to ask is not "how can it be better?", but "if it were really good, what would it actually do?".