Speech:Spring 2016 Brian Doehner Log


 * Home
 * Semesters
 * Spring 2016
 * Proposal
 * Report
 * Information - General Project Information
 * Experiments - List of speech experiments

Week Ending February 9, 2016
<2/9/16> Tried researching other Sphinx university Users, Read Logs. Watched some unit tutorials on Youtube for ideas on how to move around.
 * Task: <2/3/16> Figure out what the system is and what all of the data consists of. Try to figure out how unix works.   Start to research other Sphinx users to see their train and test configurations.

<2/9/16> Found a ton of information from Carnegie Mellon University. Trying to sift through all of information on their projects to see if it relates to ours. http://www.speech.cs.cmu.edu/ is their main Speech page.
 * Results:

<2/9/16> Continue reading through the Cargenie Mellon information to see if we can apply anything. Also check on Switchboard users as well. <2/9/16> Haven't gotten to connect from home yet. That's really a problem.
 * Plan: <2/3/16> Log in to Caesar and see where the corpus is located. See what's there.
 * Concerns: <2/3/16> Connecting to cisunix from home. The amount of directories to check.  Time.

Week Ending February 16, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 23, 2016
<2/20/16> Read Logs to check on other teams progress and understand their parts of the project. I'm trying to get as familiar as possible with all aspects of the project. <2/21/16> Read Logs. I also checked on my teams proposal to be on the same page as everyone else on the Data group team. Check out more from other groups to see progress. <2/22/16> Started downloading the correct software for this project. Get it configured.
 * Task:


 * Results: I was able to get all of the software on my machine.  Putty, SSH, SoX, Office, Pulse.  All installed.  I'm having serious trouble getting Pulse to connect to UNH.  I tried from home and work, but I still can't get it work for me.   I plan on asking my teammates for help on this situation.   One thing I tried was going to Campus and trying it from the secure network at school.   I was able to do that, but that wasn't surprising at all.  Using the .csv file, I was able to make a list of all 30 .sph files for this class.  I then logged onto Caesar.   I tried Justin's method of using Pcsp to download the .sph files.  I kept getting bad pathways for the file.  So I tried to manually ftp the files, but I couldn't get the linux commands to work.   Lastly, I tried Filezilla to connect, but that wouldn't allow me to connect to Caesar.   Not sure if that's by design or not.

I'm getting a new machine this weekend. My old PC couldn't handle the task that was required of the work. So I finally bought a new computer to make sure that I can access Caesar from home and not just using the School computers. Downloading SoX, FileZilla, and a copy of Office are the first on the list. Once I secure those, it's setup the VPN to access Caesar. Using the Excel that Justin provided has given me the list of .sph files that I need for the 31 transcripts to check against. After I copy those to my machine, I'll use SoX to convert to .wav files to use to listen to the Switchboard telephone records. Make any annotations that may be needed regarding both the .wav files and the transcripts as well. Lastly, I want to check the other groups reports to find out what's been done to all of them. Getting all of the software up and running correctly on Windows 10. It shouldn't be too hard, but I still have misgivings on if it will all work. 30 transcripts doesn't seem like much, but I'm unsure how long some of these calls really are. Getting the files from Caesar to my machine without breaking anything in Caesar. I think I have an irrational fear of breaking something in the Server.
 * Plan:
 * Concerns:

Week Ending March 1, 2016

 * Task: The Data group divided up 242 .sph files amongst the four of us.  We have to listen to the calls and check them against the transcripts.   First thing is to download the files and the transcripts.  Next is to convert the files to .wav format to be able to listen to them.   Next is to listen to them and mark any differences.  This will take some time, but shouldn't be too hard.  I also plan to re-read the proposal and check my teammates logs as well.


 * Results: I had copied and downloaded all of the files before I left class on wednesday.  If I can do that for future classes, I think that's not a bad idea.  I made it through all 60.  I didn't notice many huge errors like my teammates.   All of mine were more or less correct.  Specific problems that I noticed:

sw2507A-ms98-0099.sph - the word Control was missing.

sw2513B-ms98-0041.sph - the word Clutch is listed as being part of the laughter. Not spoken at all.

sw3517B-ms98-0006.sph - there's strange -1 after a couple of words.

Everything else seemed fine.


 * Plan: Download the files before I leave class.   I have a funeral this weekend in new Jersey.  So I'll bring my laptop to listen to the files on the trip down and back to make use of what time that I can.   Mark any deviations between the calls and the transcipts that I notice in the results area.  Listen to each one at least twice to three times to hear anything that might be wrong with the files, then check them against what the transcripts are asking.  I want to make sure that I'm as accurate as I possibly can be when checking these.


 * Concerns:  getting through all of them.   Not sure about how bad some of the problems might be.  How good is my hearing on this many?

Week Ending March 8, 2016

 * Task: Contact other universities about the Corpus to get Train and Test information that we could use.   Get and listen to more .sph files.  Mark any weirdness.


 * Results: My first 13 wav files all had the same data.  186-199 of the random sample that Justin Creates every week.    "I was, That's Exactly what I was going to ask you, have you been Downtown recent?"   The thing that's really interesting about it is that line is nowhere in our list of translations from this week.  Now I must look further to find out where the translation is located in the corpus on Caesar.   I haven't heard back from Johns Hopkins, Dragon or Berkeley on find the train and Test amounts using Switchboard.   Cleaned up all previous weeks into organized files for reference by file size.


 * Plan: Find the White paper that I gave the group from Berkeley and contact their Speech recognition department.  See who else was listed on it and contact them as well.   I know about Dragon Systems.   Run Sox on my new .sph files and listen for data correct.   Organize the .sph files on home PC for future reference and referrals on this project into folders by week.


 * Concerns: Getting anything back from the other universities.   How much bad data?

Week Ending March 22, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 29, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 5, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 12, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 19, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 26, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending May 3, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending May 10, 2016

 * Task:


 * Results:


 * Plan:


 * Concerns: