On to the Project Plan

After getting the digital mock up done and presented, I completely forgot about posting another blog. Anyway, click here to view the digital mock up. It should be processed by the time people read this post. It's basically how I think the application would work and how users would interact with it.

So the results of the presentation... I haven't checked yet :p The tutors did give us a lot of feedback since everyone had some sort of experience with either voice recognition software or early child education. One notable suggestion that could potentially be molded into our project is moving the entire application to another platform... PC based -> Nintendo DS.

It turns out that the DS already has something similar to what we're proposing. I think it was called Brain Training which is comprised of a lot of mini games, one of which being a voice training application. If DS coding isn't too complex and there are existing voice recognition libraries that are importable, the scope of the project could completely change.

After researching the development of DS games, I was informed that the general name for DS coding is "homebrew". I later came across a site called NDS HomeBrew, which as the name suggests, has a lot of support for anything DS.. especially develop. I've posted a thread on there asking about voice recognition with the DS in the hope that someone can help with it. If not, I may have to go back to using Dragon and Flash. Developing on the DS would be much much nicer :p

The thread can be found here.

Getting back to Dragon, I did some testing with it... so... setting up an account after it's been trained to your voice works to an extent. Of course it's not perfect so there are still a lot of limitations with this type of technology. Stuttering certainly doesn't help. It was a lot more rigid than I thought it would be.. for stuttering, it needs to be a little more flexible. Though I do understand that making it more flexible will probably make it less accurate.

After the initial setup, I ran through a few test cases speaking with various accents, pitch changes and tones to see how much the results would be affected. The accents I tried were: Indian, Asian (generalised and really obvious), intentionally stuttering, talking with a high/low voice and asking someone else to talk.

The results... well, some things changed. A lot of problems came up with recognising even slightly complex words when accents were concerned. Stuttering sometimes gave null results. Asking someone else to talk gave similar results to accent changes. As it stands now, training will be crucial to accuracy.

However, all of these results could actually be nulled since after I played back what I said just to test the mic, I found that the mic itself was slightly damaged and was letting in an annoying clicks and pops which may have affected the voice recognition. Until I purcahse a better mic, I won't be testing too much more since the results may not be so accurate.

I'm hoping to get a reply on the NDSHB forums ASAP so I have a little more focus and gauge if this will be possible or not. If the coding is too intense, I don't think I'll have the abilities to really do this with the DS :/