Getting towards 100%

Extensions to my voice classification project

The biggest difficulty, in my opinion, with machine learning problems is after you've got a working solution. You're staring at some accuracy figure, say 95%, and wondering 'how high can this go?' There are a huge number of options for that next step.

Neural Style Transfer is a technique for transforming an image (like this photo of the seascape in Qingdao) into the style of another image, typically a painting (here the style is of Hokusai's 'Great Wave off Kanagawa').

Who is it?

Playing with machine learning to classify voices

Someone calls you on the phone, and within a few words you can work out who it is. Can a computer do this just as accurately? For a bit of fun, and to practice some machine learning, I made to program which does this. I give the program a few minutes of recorded speech from 13 people to learn from originally, and it can tell from a few seconds of speech who is speaking with an accuracy of 95% (using Mathematica) to 97% (using Python).