Classifying transcribed speech with Sklearn