I have many google homes in my flat. I use them to turn on the lights, set timers, do unit conversions and play ocean sounds. I wouldn’t consider their speech recognition to be good. My girlfriend with her feminine voice has an even harder time with them, often resorting to putting on a comedic deeper voice to get google to recognise her commands. I’ve looked into how speech recognition is done and when building my computer back in January one of the goals I had for it was to train my own language model. I’ve written about ideas I have for getting the computer to recognise language before. My understanding of the most up-to-date techniques for language recognition is to use a CTC network to train between the audio and words. The training data that is used is labelled speech. That would consist of a mp3 file of someone saying a sentence and then a text file or record of that sentence. The audio file will then be decomposed into a frequency analysis creating an image like the following w

  If you haven’t at least heard of the book capital in the 21st century, then you haven’t been keeping up with your economics Nobel prize winners! The focus of the book is on the premise, backed up extensively throughout that the rate of return on capital has always been and without intentional redistribution by a state like entity will always be larger than the rate of growth in the economy. If you are happy to take that statement on reading it then perhaps you no longer have any need to read the book, if you are sceptical or at least intrigued in the development and evidence for the concept then you can get the book on audible. One question that isn’t addressed at least in Tomas’ first book (I feel he addresses it better in “capital and ideology”) is what stops this from being the case in the limit. Surely you can’t go on earning a larger and larger share of the pie. There is after all only so much pie to go around. It is a relatively simple calculation to work out for a given long t

 I've had my oculus quest for some months now and I've been thinking about how I can make the best sunset experience possible. One of the ideas that I was exploring was that of prerendering. Since in a sunset experience, I don't expect the user to be able to move much out of their drawn area, there is going to be a maximum distance for which they can discern parallax. Using the area I can draw out, a 2m x 2m square, we can calculate how far away in virtual space an object needs to be before its motion from parallax would be sub-pixel. A      B  |\     |  | \    |  |  \   |  |   \  |  |    \ |  |     \|  C      D Using the above diagram, we need to calculate at what distance The oculus quest has a pixel density of 20 pixels per degree. That means that each pixel is 0.05 degrees in the field of view. From that we need to calculate at what distance away from the user does an object have to be before the change in angle is less than this one pixel. That is if the user moves fro

 It is so often an argument, does calorie counting work, is a calory of sugar more fat inducing than a calory of celery. I’ve never really understood why one source would be more energy inducing than another given that they have the same amount of energy. Well, I’d like in this post to present what I have found recently starting with sugar. When learning biology back in my GCSEs we were taught that the body uses glucose as a fuel and that glucose was sugar. Now although I’m sure the commutativity of that equality was never explicitly confirmed I had presumed that if glucose was sugar then sugar must be glucose right? Kind of. Sugar is made of a disaccharide called sucrose, this is a molecule made up of one glucose and one fructose. The body starts breaking this down in the saliva and then more so in the small intestine, where the glucose and fructose molecules are absorbed into the blood. Now unlike glucose, most of the cells in the body can’t use fructose as fuel. That means that the

I have been thinking a lot recently about relativity (Started drafting this post over a year ago). After having made my last post on black holes I have been watching a lot more on general relativity as well as reading lost in Math and The Fabric of the cosmos. The only reason I can consume so much is because audible allows me to listen at up to 3.5x which when played out loud feels very much like the information downloading from the matrix. One comment from this video was that as a result of special relativity, light trapped in a perfectly reflective massless box will act as if it has mass. I haven’t been able to explore this in enough detail to truly understand it but it seems that mass can be an emergent property of the energy. Clearly, this is shown in e=mc^2 but the idea that the mass form of the energy can emerge in the first place.  The other train of thought I have been playing with is the idea that light is the propagation of causality and that causality itself is time. That m

In a previous post I've solved issues of a farmer trying to maximise the area of their field using a given length of fence. Given that was solvable I thought, well what if rather than a wall the farmer was to use a circular lake? This is also modelable but the answer isn’t as intuitive. I would argue the reason the answer isn’t as intuitive is that the more of the fence circle you use, the more the lake invades the pasture. There must be some benefit in using the edge of the lake though, at least my intuition leads me to believe. Here I try to go about calculating the systems in the same way as in the previous post. Let the reflex angle BAD represents the large simple area of the pasture. The segment ABE represents half the remaining segment, bearing in mind that not all of this is usable. The area BEFB is subtracted from this and represents the area covered by water that would otherwise make the remainder of the arc.   This can be calculated i