I looked into it two month ago I think and asked for the sound recording again.
Also I singed in several free voice recognition API's out there and tried my best.
I thought that it could be possible to dictate into a hidden Editbox with focus or as I mentioned somewhere to use another script to record and making the request as we even can't change the header for the HTTP packets.
But I managed to make a separate script which does the recording and gives the right text back.

And I tried to create my own HTTP packets via the socked commands but as long as we cant record sound we rely on a second script/program.
Ah here it is ...I used curl.exe to make the requests and the ffmpeg.exe to record sounds then I wrote batch which creates a JSON file with the needed data to send to the google speech API... the rest did Jack mention before

Just ask If you want the files