My 'dodgy' microphone.
dodgymic.wav
My voice recorded using the dodgy microphone.
Voice Recognition
With that out of the way, it was time for something different;
voice recognition. The Andrea Pro Gamers include a noise-cancelling
microphone. While I'm sure that there are technical definitions
on how this works, I preferred the practical approach. First I
recorded my voice a few times, and was pleasantly surprised at
how clear it sounded – especially compared to the other 'dodgy'
microphones that I'd used before. Given the old saying "a picture tells a thousand words", I'm not
sure how many words sounds count for, but here is the experiment:
This first .wav file was made using my 'dodgy' microphone; this microphone was supplied free with some of the Dell or Compaq computers we got at work several years ago. As you can hear, there is significant static buzz as well as lots of interference from breathing too close to the microphone while talking.
The second .wav file was taken using the Andrea 1212's noise-cancelling microphone. As you can tell, the voice quality is significantly better, there is very little interference and it is easy to tell why any voice dictation software would function much better with the noise cancelling technology. I also found that the 'Mic Boost (+20dB)' option – in the Windows Volume Controls - made voice recording much easier.
Installing the Dragon Preferred V6 software was quite simple, even if it did force a reboot upon completion. After starting the program for the first time, I was able to select Australian English, and start a few simple 'training exercises'. The training exercises involved reading through several pieces of text, so that the recognition software could recognise my specific voice patterns. I expected to spend at least a few days 'training' the voice recognition software before it would be usable to any degree. However after only about 15mins to half-an-hour, I was able to jump on IRC and try it out. As you can see in this log, not everything turned out as it should, but on the whole, and given that I'd only been using the software for an hour or so, I thought the results were pretty good.
After a bit of play on IRC, it was time for some more training, to see if I could get even better results from the voice dictation software. I had already worked out how to add specific words to the vocabulary, like people's names and other words that it did not want to recognise. The additional training simply involved reading a few more passages of text, including JFK's inauguration speech and some chapters from a few books.
I felt the dictation software was working well enough to attempt to write the review with it. Using Dragon's simple 'popup help' and the Andrea Pro Gamer, I was able to write the entire first page of this review, without any major dramas, and more importantly without needing to use the keyboard or mouse! (The first-draft output, completed using only voice dictation and voice commands is linked below.) I did need to use the 'undo that', 'back a word', 'forward a word', 'select that', and 'delete' commands often, but I didn't need to use the mouse or keyboard for the entire document. Given that I had only been using the Dragon Preferred software for half-a-day, I thought this was quite impressive.
I was having a bit of fun 'talking' on mIRC, or any other program that I had open, I could also use the software to switch between programs, open new programs and perform a variety of other tasks. Even more fun was the ability to teach the Dragon software a collection of new words; woot, hehe and Wolfy are some that I can print here, it did refuse to learn a few words however, such as Agg, and Soc, but that was not really surprising given they are nicknames rather than real words.
One other thing that I noticed when using the headphones and dictation software, was that my usually appalling spelling was improved significantly. When using the dictation software and voice commands I did need to use the 'Select that' – 'Choose X' options to change words that I had dictated, like wear and where as well as correcting syntax and other errors. The program was also very good at 'natural language' recognition, in that it generally performed better when entire sentences were dictated using 'normal speech'. While using 'robotic talk' to dictate one word at a time the software tended to interpret words and syntax incorrectly. The Dragon voice recognition software was also smart enough to filter out 'non words', like ummm, ahhhh, and errr, and on the whole I found it surprisingly easy to use.
 DIY Parallel-Cable Remote LCD! |
 TwinMOS PC2700 DDR |
 MSI K7N420 Pro nForce Motherboard! |