Speech:Spring 2015 Garret Bryant Log


 * Home
 * Semesters
 * Spring 2015
 * Proposal
 * Report
 * Information - General Project Information
 * Experiments - List of speech experiments

Week Ending February 3, 2015

 * Task:
 * 1/27
 * Met with group members and discussed future plans for project. Established means for communication between the group.
 * 1/31
 * Read through logs to familiarize myself with the project at hand.
 * 2/3
 * Read more logs.


 * Results:


 * Plan:
 * I plan to learn as much as possible about what I will be doing throughout the rest of the semester.


 * Concerns:
 * I fear that my lack of knowledge about the project may make it difficult to read other people's logs about it.

Week Ending February 10, 2015

 * Task:
 * 2/7
 * Read logs and reviewed scripts.
 * 2/8
 * Read more logs.
 * 2/9
 * Attempted to log into Caesar from home on my personal computer. Log in was successful and I was able to view the directories without any problems.
 * I practiced several commands to help navigate through the interface.
 * I started constructing a list of commands that may come in handy in the future.
 * If I feel that I have constructed a useful list tomorrow then I will post it to the results section of this log.
 * 2/10
 * I have walked through all of the necessary steps that I will need to run a train.
 * I intended on performing a test train tonight, but my home internet connection has been very inconsistent so I will postpone it until tomorrow.
 * I created a table of the commands that I have collected, but it didn't submit properly, because of the connection issues.


 * Results:


 * Plan:
 * I will have to work on my plan for my contribution toward the modeling group.


 * Concerns:

Week Ending February 17, 2015

 * Task:
 * 2/11
 * Ran train with Sam and Zach.
 * Worked with group to establish individual timelines.
 * 2/13
 * Read logs.
 * 2/15
 * Read through logs and other information on the wiki to determine how to fix an error while trying to run the decoder.
 * 2/17
 * Discussed the issues with decoding with Sam and Zach.
 * Sam seems to have found a solution to that
 * Now we need to figure out how to change the values when running a train(current method uses default values only)


 * Results:
 * We now know how to run a train and decode it.


 * Plan:
 * Run trains to test different senone values
 * Concerns:
 * We encountered a problem when trying to run the decoder.
 * Following the tutorial just spits out an error.
 * It seems that we will have go about this in a more manual way.

Week Ending February 24, 2015

 * Task:
 * 2/21
 * Read Logs
 * Adjusted our group's task breakdown on the proposal with Zach
 * 2/22
 * Read Logs
 * 2/23
 * Review existing tutorials to find unnecessary overlap
 * Sketched out a plan of how to organize the future layout without entirely deleting the old tutorials
 * Researched more in the scripts on how to adjust values properly
 * 2/24
 * Discussed with Zach on what our next steps will be
 * Commenced further planning on the documentation for running trains
 * Also planned more on the trains that I will be performing


 * Results:


 * Plan:
 * Talk with Sam on what the status is with trains
 * Concerns:
 * Communication has been bad between our group and we will have to try to improve this.

Week Ending March 3, 2015

 * Task:
 * 2/26
 * Compared pruneDictionary scripts to look for differences between versions.
 * Made modifications to the latest version to allow it to work properly.
 * 3/1
 * Read logs
 * 3/2
 * Set up the 125hr_clean train for experiment number 0266
 * Started to run the train, but I ran into an issue with, seemingly every word, not appearing in the dictionary
 * 3/3
 * Met with team about our progress.
 * Found out that the train that I was doing might not have been prepared properly
 * Cleaned out the failed part of my last experiment and prepared to run a 5hr train that should work without any issues.


 * Results:


 * Plan:


 * Concerns:

Week Ending March 10, 2015

 * Task:
 * 3/4
 * Have to wait for data group to finish changing things before more trains can be run
 * Fixed the Experiments page so the Expand and Collapse button line up with the title properly
 * 3/9
 * Met with Zach to attempt to fix the issues that we encountered with the genTrans11.pl script.
 * Changed the line of code that establishes the softlink with the utt folder in switchboard
 * the directories recently changed, so we had to make it: .../full/train/audio/... instead of .../full/audio/...
 * When signing into Caesar there was an issue that brought me to Brutus instead.
 * Spoke with Mohamad and the Systems group is working on the issue
 * The systems group will also be migrating to Brutus at 5pm so I have to wait to do anything else
 * 3/10
 * Checking in


 * Results:


 * Plan:
 * 3/8
 * Will attempt to run another 125hr train
 * Concerns:

Week Ending March 24, 2015

 * Task:
 * 3/18
 * Hosted a meeting with the team
 * There was not a good turnout of people, so we will have to do our planning through email
 * Caesar is down so there was not much that we could work on together
 * 3/21
 * Read logs
 * Caesar is still down...
 * 3/22
 * Read logs
 * Still can't do anything on Caesar
 * 3/24
 * Professor Jonas said that Caesar back up
 * I tried logging in, but I am still getting a "The host is unreachable." error
 * It is a little late to ask for help tonight, but I will try it again in the morning


 * Results:
 * With the server being down for a week, there wasn't much to accomplish.


 * Plan:
 * 3/12 - 3/18
 * Organized meeting with group and passed along information about what we should be doing in the coming week.
 * Concerns:

Week Ending March 31, 2015

 * Task:
 * Examine the scripts that I have been tasked with as seen in our group page
 * Become more familiar with running a train


 * Results:
 * 3/27
 * I gave the people who missed class a complete update on what we did as a team, and what they should be doing before next class
 * I worked with Dakota and Nicholas to try to solve an issue with decoding
 * When executing run_decode5.pl we were getting a does not exist error
 * Going into the base experiment directory and running DECODE/run_decode5.pl seems to fix this issue
 * I did have to move the decode.log into the DECODE directory afterwords (mv decode.log DECODE)
 * Ran a 5hr train to make sure that everything is working properly (0266/003)
 * There were no problems that I encountered except for the one described above
 * I got a Sum/Avg Error rating of 42.2% with all default values (including 1000 senone value)
 * 3/29
 * Examined two of the four scripts that were assigned to me by the group. Details are on the group page.
 * Went through several directories and files in a basic experiment to see what I could modify for better results.
 * Ran another 5hr train, but with slight modifications this time (0275/001)
 * I got a Sum/Avg Error rating of 39.3% this time
 * This is about a 3% improvement which is good, but not enough
 * More details are on the group page
 * 3/30
 * Read through logs to see teams progress
 * Tried to log into Caesar but ran into a problem
 * Permission denied, please try again.
 * It's getting late, so I'll have to ask around tomorrow to find out what's going on
 * 3/31
 * Read more logs and last year's experiment results
 * Still can't log into Caesar even as root


 * Plan:
 * I plan on stepping it up to the larger trains soon (125hr or maybe even 256hr)
 * Concerns:

Week Ending April 7, 2015

 * Task:
 * 4/1
 * Tried to run a 125hr train several times
 * Ran into errors with the dictionary missing words
 * After adding those words, verifyAll.pl was still giving off an error with no explanation
 * I noticed that the number of hours of audio on the log said it was less than 6 hours.
 * Instead I started a 256hr train to try to get some kind of results with that
 * The details of this experiment will be posted on the group page when we get the secrecy of the wiki under control
 * 4/4
 * Read logs
 * Still waiting on the 125hr train to finish
 * 4/5
 * Read more logs
 * Train is still going
 * 4/6
 * The train finished running, so I created the language model and started decoding with 5000 audio files
 * There are a lot of errors that occurred throughout at least 3 of the steps in RunAll.pl
 * I noticed them in script 20... 30... and 50...
 * An example of the errors: ERROR: "baum_welch.c", line 331: sw2884B-ms98-a-0135 ignored
 * 4/7
 * I got results for the 256hr train
 * The Error rate was 41.2%
 * More details will be posted to the group page


 * Results:


 * Plan:


 * Concerns:
 * 4/1
 * The audio files in 125hr train were all broken soft links, which is what was causing errors.

Week Ending April 14, 2015

 * Task:
 * 4/11
 * Read Logs
 * 4/12
 * Read Logs and did some sphinx research
 * 4/13
 * Did more research
 * I found out what is causing the errors on training, and I will post more on the group page
 * I also discovered some potential solutions to getting a better language model in sphinx


 * Results:


 * Plan:


 * Concerns:

Week Ending April 21, 2015

 * Task:
 * 4/17
 * Attempted to start another 256hr train
 * I was almost able to type out the first script name in 30 minutes
 * The wifi in this hotel is awful
 * I did end up figuring out exactly what I will be changing to get better results
 * 4/19
 * Tried to do another train, but it said that "something failed" during verify stage
 * Couldn't find anything wrong with the logs, so I just retried it
 * I'll have to wait until the morning to see if everything worked properly
 * 4/20
 * Train appears to be running successfully. Not sure what was wrong last night.
 * Details regarding the experiment will be shared with the group
 * 4/21
 * Train is still running
 * Read logs


 * Results:


 * Plan:
 * Begin a 256hr train when I have more than 10B/s internet speed


 * Concerns:

Week Ending April 28, 2015

 * Task:
 * 4/24
 * Still waiting for the train to finish
 * Re-ran a decode for my earlier 256hr train to try to get better results, but it did not seem to change much.
 * 4/26
 * The train is still running, so hopefully the time is worth the results
 * Read logs
 * 4/29
 * Train finally finished today, which is why I haven't been able to do anything this week.
 * I had been communicating with team members to fix some problems.


 * Results:


 * Plan:


 * Concerns:

Week Ending May 5, 2015

 * Task:
 * 4/30
 * Created a script that may help with results. Script and purpose has been shared with team.
 * Started decoding my latest 256hr train
 * Re-decoding my first 256hr train on a drone for real time factor
 * 5/1
 * The re-decode for the old 256hr train finished
 * The results have improved slightly from some modifications discussed with the team
 * The newest 256hr train is still decoding...
 * 5/2
 * Previous train finished decoding
 * The results were very promising and have been shared with the team
 * 5/5
 * Real-time factor has been captured


 * Results:


 * Plan:


 * Concerns: