I am using a 10gb expanding partition, with 512Mb of RAM allocated to the VM.
If I give it any less then 512mb of RAM the recognition accuracy gets really bad. Also If I give it more then 1GB of RAM (in my Macbook with 2GB of RAM) then it also loses accuracy, due to not enough RAM on the Mac side.
I ahve not tried diffrent partition sizes, but I have found that as long as the partition does not change size much (i.e. creating large files) Dragon works as well as it does on my real hardware Windows box. (I am using the same Dragon user with the same USB mic to hold things as constant as I can)
I changed the sound input in the Mac side. The audio driver inside Windows seems to mirror whatever the Mac gets as input, regardless of what is selected in it.
I dictate to Dragon pad, and save the files in a SMB share so I can get at them directly on Mac. The only down side is that Windows likes to have ^M's at the end of each line, and that Dragon Pad does not add carrage returns at the end of lines when the wrap. (This is plain text files)
I'm working on a little python script that will fix the second one. Fixing the first one is simply calling dos2unix on the file.
Also, I do not get a CPU spike unless I tell Audacity to record. If I let it "monitor input" by clicking on the VU meter on the right, then minimize the window and ignore it then it does not eat my CPU for lunch.
Sometimes if I suspend the machine while Dragon is running, then resume it without the mic plugged in (or even with the mic plugged in, but not selected as the input device), then Windows sound support is borked completely until I suspend/enable mic/resume again. This is annoying, but whatever, I can talk to my Mac now
Last edited: Jun 16, 2006