Voice recognition with the first Kinect was far from perfect, but Microsoft promises that things will be much improved with Xbox One.
Today we learn that one of the problems that affected the audio design of the original device was bound to the personnel involved, as explained by Microsoft Research Principal Software architect Ivan Tashev in an extremely interesting 20 minutes long interview on Channel 9.
I still believe the Xbox team was, and is, one of the best engineering teams we have at Microsoft. The problem I had to face was that those are enormously highly trained software engineers. All the procedures for creating, debugging and building the code are established and followed strictly.
But not one of them had, for Xbox’s Kinect, experience in doing audio signal processing code. Which means that they had difficulties to find a problem when it is not a coding problem, when it’s an algorythmic problem.
And this is what basically forced me to go and to spend the four months before Xbox’s Kinect’s release there. I had an office there, I had a desk there, I was working hand by hand and soulder by shoulder with the software engineers.
What we tried to do for the Xbox One is to bring people with the proper qualifications, so now the audio team in Xbox consists of four people with PHDs in signal processing, so for me it’s way easier to work on new algorythms and to hand over the MATLAB scripts, and they can convert and debug and solve most of the problems which are specific for this particular implementation.
We can only hope that the new engineers led by Tashev managed to nail audio recognition this time around, as it will be critical to many of the Xbox One’s main features. In the meanwhile you can watch the full interview below. It’s extremely interesting and it’s very much worth twenty minutes of your time.
As a bonus, it also shows the back of one of those lovely white Xbox One units, that you can see at the bottom of the post.