Author: Zachary Dudek
Purpose: First Experiment
Results: The results of all of the experiments are detailed below. This entry will consist of the overall detail about running the experiments on Caesar, Asterix, Miraculix, Obelix, Majestix, and Idefix. Essentially the goal here is to make sure the word error rates and all of the other statistical results from running the experiments are similar, if not the same across all of the drone machines compared to the result of Caesar. The reason for conducting these experiments was to make sure that the software installations on the Caesar and the drone machines were not going to be skewing the results of the data from the experiments. From running the trains, creating the Language models, and running the decodes on each of the drone machines and Caesar I can say that the results are mostly identical on the machines that were able to run the scripts in full. Miraculix and Majestix ran in to some hurdles and I was not able to get them to run in full. I will detail some of what happened and possible fixes for those two drone machines in the individual experiments below. The common word error percentage was 43.7 on the drones will running a standard 30hr train and decode from the corpus.