The other day, I saw that a version of MyCroft was released for the Raspberry Pi. I have been following MyCroft for a while now (mostly through the Linux Action Show) and have tried using it. The software is still in beta, so I found some bugs with it, namely I can’t really use it. I have tried pairing it, then talking into it with my microphone. But it can’t understand what I’m saying. At first, I thought it had something to do with getting a secure connection to the MyCroft backend servers. Now it could be a problem with my microphone.
I’ll admit that my desktop microphone isn’t the best. But how much clarity does the microphone require? Apparently, a lot. The microphones on the Amazon Echo, for instance, can pick up a bunch of channels of sound. So it looks like I’m going to have to get a better microphone.
What I’ve also seen is that MyCroft uses Google’s backend for the speech recognition. It looks like they’ll go to something such as Kaldi, but that doesn’t have a large enough speech model to get the job done. While it has a model based upon over a thousand hours of speech, it may require thousands of more hours of speech just to get better results. I’ve been donating to Voxforge and trying to help with their speech corpus. However, they’ve barely got enough for half of their first release. So I was wondering how to speed things up and get them more samples.
What they could do is make it fun and interesting to donate. I was thinking of something like a badge system on the Voxforge website, or even leaderboards. Then again, would this make it fun to donate? I need to think more on this.