Land of Gregs

This is just a quick entry to explain why I haven’t been blogging recently. The simplest explanation is that I had a chest cold. I have a number of conditions that make this rather more serious than for healthy people, but that’s no excuse for not blogging! The real reason is that I use speech recognition when I can’t use my hands. As you can imagine, given the current state of speech recognition, a chest cold is not a good thing for getting a computer to understand you. In fact:
“It can imagine giving them a better holding more than they are getting your computer to the new room and”
was the output of my computer’s speech recognition to the previous sentence with my cold.
Speech recognition typically uses a form of a mathematical model that I have discussed before – the Markov model. In fact it uses a more complicated version of it called a Hidden Markov Model or HMM. This uses a set of Markov nodes(states that have a probability of linked to other states) that are not connected to the input or output (and so are called hidden) to allow the model to have a form of “memory” that simple Markov chains don’t. Thus, they can be used to recognize what is likely to come next while the input is streaming in, and so recognize speech. The trouble is, if the input is really different to what it’s used to, it messes up the probabilities, and you get a sentence a bit like the one above.
Hopefully normal service will resume shortly when either my voice or my hands are more usable. Without voice recognition, even this short entry took a few sessions to complete