IVV, I have used voice command exclusively for about a year now. I run Win7 without SDK and I ran through the voice training module maybe twice IIRC. I have very few problems with misrecognition. I use a Sony PS wireless headset, but I've also used a cheapo wired one from the Shack. Both give error rates about 1-2% - that is, there will be a "doink" - system did not recognize command - once every 50 commands or so. I get misrecognitions that trigger a Windows command less than once per gaming session. I tried an expensive Logitech headset with great 7.1 sound, and it gave error rates of maybe 10% and would not recognize some words - like "Fire!" - at all, ever. Tossed it. So reliable voice recognition seems to depend on having a good microphone, and "good" does not equal "spendy".
I have used Eric Knepfler's shspeech software for SH3 and SH4 and the built in voice command capability for SH5. I found setting the confidence level in SH5 to .85 worked better than the default .70. There is no similar setting for shspeech. (Don't know about Win8.) I have pruned the voice_commands.csv file pretty ruthlessly. I try to limit myself to just two or three different phrases for each command. That helps narrow down the choices the software has to make. R/L there were standard ways of phrasing orders to avoid confusion, so that is appropriate. You can throw in a couple of favorites like "Take 'er down." but you shouldn't have ten ways to order ten knots.
