Speech:Exps 0283

Description
Author: Jon Shallow Date: 2/4/16

Purpose: Reproduce last year's score

Details: Duplicate the best results from SP 15' using 125hr_train.

Results:
 * 001 Initial Train
 * 002 Second Train
 * 003 Third Train 125 Hr -- Duplicate Best Result from Spring 2015
 * 004 Fourth Train 125 Hr -- Bring convergence ratio to 0.001, run variance normalization
 * 005 Fifth Train 125 Hr -- Increase dictionary size limit to 30k
 * 006 Sixth Train 30 Hr -- Test the new "fixed_30k" corpus
 * 007 Seventh Train 30 Hr -- Test the new "fixed_30k" corpus
 * 008 Eight Train 30 Hr -- Test the new "fixed_30k" corpus
 * 009 Ninth Train 4hr -- Test the new "corpus"
 * 010 Tenth Train first_5hr -- Test the new utterances in /full/audio/utt/
 * 011 Eleventh Train first_10_lines -- Test the new utterances in /mnt/main/home/sp16/jrs1036/first_10_lines/train/audio/utt/
 * 012 Twelfth Train -- Create new experiment to verify decode leg of process
 * 013 Thirteenth Train -- Test the newly generated utt files after fixing the genUttAudio script
 * 014 Fourteenth Train -- Testing new 150hr corpus with default values on new utt files
 * 015 Fifteenth Train -- Testing the new utt files (sample_n_bytes = 4)
 * 016 145 hour train -- Testing the new utt files with 32 density and 8000 senomes(sample_n_bytes = 4)
 * 017 145hr train -- Replicating 016 to verify new emacs install doesn't interfere with dependencies
 * 018 145hr train -- Replicating 016 to verify new emacs install doesn't interfere with dependencies -- Take 2
 * 019 30hr train -- Determine if s tags are necessary for training
 * 020 30hr train -- Create a baseline to compare 019 to