Speech:Spring 2012 April 24th Group 5

From Openitware
Jump to: navigation, search


Sub Groups


Group Members


Group Log

Experiment number: 0005

Machine: Verleihnix

  • Matt
  • This week we worked on performing the steps to run a train found here. Since our system is no longer setup the way it was last summer when these directions were written, we've had to accommodate by making several changes. This is mainly due to the fact that the shared directory /mnt/ (where critical files are now stored) will be executed locally on batch queue machines. This wasn't the case during the summer of 2011 when the original instructions for training and decoding were written, so we've had to adjust accordingly. If you are using these notes as a guide to perform the train, I suggest having the summer training instructions open and I will refer to each section by subheading name and step number.
    • The first problem we ran into was that setup_SphinxTrain.pl in step number 3 subheading "setup task directory" was looking for a directory relative to it's previous path. Since we moved the file to /mnt/main/Exp/0005 (our particular experiment number), it was no longer able to located the files needed while it was on caesar. Professor Jonas helped us open the file and find the variable which contained the path, and we hard coded it to point to /mnt/main/root/tools/SphinxTrain-1.0/scripts_pl/setup_SphinxTrain.pl which should essentially work for all experiments located in /mnt/main/Exp/<experiment name>. The name of the variable is $SPHINXTRAINDIR.
    • Step 4 under "setup task directory" requests that you copy sphinx_train.cfg but does not specify a directory. The cp command has 2 parameters:the first is the path of the file you are copying, the second is the path of where you are going to copy the file. If you don't provide a second argument, the cp command should place the file in the location in which you are currently located. This doesn't seem to be the case for our kernel, however, and we had to specify the second argument in order to get this to work. Group 1 was working on this step simultaneously, so they copied the config file from /root/DOCS/ to /mnt/main/root/docs/ using the following command: cp -i /root/DOCS/sphinx_train.cfg /mnt/main/root/docs
    • Step 1 of the "Copy wav files into wavTemp directory" subheading remained the same. We created the directory /mnt/main/Exp/0005/wavTemp in preparation for the following steps using the command: mkdir /mnt/main/Exp/0005/wavTemp.
    • Steps 2 and 3 of this subheading don't work because once again the second parameter for the copy command isn't explicitly named. The command we used: cp -i /media/data/Switchboard/disk1/swb1/sw02001.sph /mnt/main/Exp/0005/wavTemp , where 0005 is our particular experiment number.
    • Next, we did the steps under the "Copy necessary scripts into the etc directory" subheading. This was fairly straightforward but again we had to explicitly name the second argument of the cp command. The commands we used were: cp -i /root/SCRIPTS/genPhones.csh /mnt/main/Exp/0005/etc and cp -i /root/SCRIPTS/genTrans.csh /mnt/main/Exp/0005/etc
    • Finally we copied the dictionary from /root/DOCS/cmudict.0.6d to our etc directory in our experiment directory using this command: cp -i /root/DOCS/cmudict.0.6d /mnt/main/Exp/0005/etc. This is as far as I was able to get. I think the next step involves taking a portion of a transcript and copying it into the /etc directory within your experiment folder. I wasn't sure how to go about getting a portion of this file into a new file. I know that the "head" and "tail" commands can save part of a text file, but I couldn't figure out how to specify a number of lines, or which transcript to use to begin with.