Speech:Exps 0185

From Openitware
Jump to: navigation, search


Author: Colby Johnson

Date: 20Feb2014

Purpose: This train is an attempt to run a train using what we know now to optimize results. Also to build a master dictionary that matches all of the full set of data. This will make running trains more accelerated. By removing a bulk of unnecessary words. The pruneDictionary script runs much quicker.

Details: This Train is run using the following parameters:

  • Corpus: first _5hr train
  • Sphinx_train.cfg:
    • Senone value: 8000
    • Density: 64
  • Dictionary:
    • /mnt/main/corpus/dist/custom/308hr.dic (Does not exist yet)
  • GenTrans: genTrans5.pl

Results Running genTrans5.pl took about 5 hours. This file needs to be looked at for efficiency

  • alternatively we could generate these setups for all corpus subsets

Currently building dictionary....

(3/17/2014)-Colby Johnson
A lot has changed since this was created. I will be remaking this Experiment in the future pending a few results currently being generated. A new take on genTrans and the Dictionary have changed the process of running this train. This will be recreated at a later date.