Speech:Spring 2013 Drew Mather Log


 * Home
 * Semesters
 * Spring 2013
 * Proposal
 * Report

Week Ending February 5th, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 12, 2013
Learn how to do a train
 * Task:

2/7/13 - Worked with team via Google Hangout and went through running a train.
 * Results:


 * Plan:

There were some initial issues with running the train initially and there are also some issues with running the decode to verify results.
 * Concerns:

Week Ending February 19, 2013

 * Task:
 * 02/15/2013: Read through old logs and experiments. Focusing on the last experiment and the guide for how to run a train.
 * 02/17/2013: Read through group logs for this semester, Focusing on the steps Eric used in successfully running a train.
 * 02/19/2013: Attempt to run a train following Eric's log.


 * Results:
 * 02/19/2013: Unable to connect to Caesar. I am unsure of whether this is due to trying to connect through the UNH Durham campus or if the server is currently offline. I was able to learn a fair bit from going through the logs. This should help when I am able to connect to the server to attempt running the train.


 * Plan:
 * Check with group to see if everyone is having difficulty connecting to the server currently. Continue trying to run a successful train.
 * Concerns:
 * With being unable to connect to the server, I am unable to gain experience in running trains and trying to figure out how to run a decode.

Week Ending February 26, 2013

 * Task:
 * 02/21/2013: Run a train to benchmark times on Obelix.
 * 02/23/2013: Read group logs.
 * 02/26/2013: Run a train to benchmark times on Obelix now that connection is working.


 * Results:
 * 02/21/2013: After multiple attempts, I was unable to log onto Obelix. It was requesting that I enter a password, which it shouldnt have been doing, and also rejecting any password I would give it. I am unsure of what the issue was but it seems to have been resolved.
 * 02/26/2013: Now that the issue with my log is resolved, I have run a tiny train for the Acoustic and Language Models to have a benchmark on approximately how long it should take as to provide a basic test for testing system stability and performance after any changes or updates. The total run time of the genTrans.pl, pruneDictionary.pl, makeFeats.pl, and RunAll.pl scripts was approximately 40 minutes. the total runtime for the ParseTranscrpit.perl and the lm_create.pl scripts was less than a minute. These times should provide us with a basic guideline with how the system is performing so that we can provide feedback to the other groups to help them determine if their updates or changes have affected anything, both positive and negative.

The next step is to get a benchmark on how long a decode will take to run. Eric has had success with running the decode, so using his knowledge we should be able to get a more solid time on how long it should take.
 * Plan:

From what I understand, the decode takes about 5-6 hours, which will make it difficult for someone to monitor constantly to provide accurate run times. Perhaps there is a way we could change the scripts to provide us with the time they finish as to avoid missing those times and skewing the test results.
 * Concerns:

Week Ending March 5, 2013

 * Task:
 * 03/01/2013 - Read logs to gain understanding of the groups progress
 * 03/05/2013 - Read logs to see where people are at the end of the week.


 * Results:
 * Next step seems to be benchmarking the time for a decode.


 * Plan:
 * I need to speak with the group about some things involving decoding to make sure I understand it properly. We make need some scripts to make it more time effective to benchmark a decode.
 * Concerns:
 * From what I have read, a decode can take over 5 hours. I need to speak with the group to see if anyone has any thoughts on how to make timing this more effective.

Week Ending March 12, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 26, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 2, 2013
Work with sub-group to complete running training and decoding.
 * Task:

03/27/2013: Met with group during class, and attempted running trains using mini/train corpus. This was causing errors that We had not seen before, causing the training to break in module 45 and terminate in 50.
 * Results:

03/28/2013: Met with group to work on running the trains and do a decode. The errors encountered with running the trains was solved towards the end of the time we were able to meet for and 2 people in the group were able to successfully complete them. The other 2 members ran their trains from home successfully

03/30/2013: Read logs

04/01/2013: Go through group emails to ensure that I understood where the group was in running the training and decoding. All members of the group have successfully completed running a train and decode.

03/27/2013: Group planned to meet the next day to work on getting the trains to run.
 * Plan:

03/28/2013: For those who did not manage to complete a train by the end of our meeting time, they were to complete it later that evening. All members of the group were to complete a decode. We communicated by email to go over any questions or issues that we encountered. The error being caused mid way through the train seems to have been figured out now, which seems to stem from not having a stress (a 0, 1, or 2) on the end of a phone. Thanks to Eric for figuring this out.
 * Concerns:

Week Ending April 9, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 16, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 23, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 30, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending May 7, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns: