Speech:Spring 2013 Harry Dodson Log


 * Home
 * Semesters
 * Spring 2013
 * Proposal
 * Report

Week Ending February 5th, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 12, 2013
See about upgrading all software except sphinx 3.7
 * Task:

2/10 - Worked with Mint 64bit to get everything running. Right away when trying to 'make' sphinx it gave "fatal error: libutil.h: No such file or directory" error. Looking for a fix.
 * Results:

2/11 - Switched to 32bit after reading that it may have the file i needed, it didn't. Searched the repositories for everything related to sphinx and downloaded all libraries and got past that error which let me get new errors. Getting too many errors to list.

2-12 - Found a tutorial and was able to install sphinx3 and sphinxtrain and they seem to be working. I am still trying to figure out how to work it on my own project aside from the tutorial. Once I get more into it I will update to the new version and confirm if things still work correctly after the update.

My plan was to get 3.7 to work with the original software on my laptop, to copy what they have now. Then i wanted to figure out how it all worked together. Then upgrade everything other than sphinx 3.7 and confirm it still works correctly after the update. Then update everything on Caesar. I don't want to update Caesar, without confirming that the updates won't damage the current system.
 * Plan:
 * Concerns:

Week Ending February 19, 2013
1. search through notes on previous semesters and compile useful documents and info 2. find any previous perl scripts for sphinx 3. ask if we can find images of Caesar to make our own virtual machine
 * Task:

2/16 - Already requested an image of Caesar if they can recover it. need to follow up - Making a virtual machine using the directions given by a previous semester did not work and are unclear. The installation failed. I was following the directions here
 * Results:

2/17-2/18 I was reading through 2011 wiki and logs to see if I could find anything of use for this year.

2011 links: Scripts scripts in personal logs JamesNickBrian, decode at the bottom A how to training models setup.

2/19 Looking through 2012 wiki to see if I could find anything of use. I found out that it looks like the scripts they were using are part of sphinxtrain, I think. Verifying an Acoustic Model Some listed I could not find though (run_decode.pl, genTrans.pl, genPhones.csh).

I may have made a huge find, installation instructions on everything Speech install. All the links in the project notes may be useful. I am adding this link in case I don't find it again summer 2012

A couple of scripts continue to come up and I can't find; genTrans.pl, genPhones.csh. I think run_decode.pl may now be s3decode.pl but I have no way to confirm this. Nevermind I just found run_decode.pl along with other scripts Notes section including modified genTrans.pl

I was hoping to find information on installing/running sphinx that would make our lives easier. I did find some good info but not nearly what I was hoping for. From reading the previous years entries it seems like we are repeating the same cycle. Maybe there is something we can do to make it easier for the next group to get up to speed faster and make more headway into the project.
 * Plan:

Not part of the previous years entries but here are some good links. tutorial, sphinx documentation, More documentation. If Caesar if unable to be recovered. I can install sphinx and sphinxtrain, but I don't know how about the other software and how to get them to work together.
 * Concerns:

Week Ending February 26, 2013
Learn how to run experiments
 * Task:

3/22 - Had a meeting on Google hangout. Josh went through running an experiment with us.
 * Results:

3/24 - Read some more about modeling. Went through the same directions we went through in the meeting Friday. Might try an experiment tomorrow.

3/26 - I worked on the virtual machine and getting 4.2 to work. I think I found my problem. When I tried to use export for the directories of ant and java, and the path of both I did something wrong. But I also think I might not even need those steps, pretty sure I don't and will test it within these next couple of days.

Learn what I can about running experiments. Work on the virtual machine. I can follow the directions but I'm not sure if I have a complete understanding of everything.
 * Plan:
 * Concerns:

Week Ending March 5, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 12, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 26, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 2, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending April 9, 2013
Create an acoustic and language model using a 5 hour section of the switchboard with group B.
 * Task:

4/3 - Went through and did an entire experiment 0066.
 * Results:

4/5 - Read through group B & C group folder looking for our individual assignments. They are not up yet but it is still early, I will check back later today.

4/8 - Did my assigned part of the group experiment, or what was my part for now, a few missing words in the dictionary. I added pronunciation to them and sent the list to Eric to add to the dictionary. There is a lot on the board for things to get done for the experiment, I suppose I will wait for my next set of orders to come in.

4/9 - Looked for more assignments, there were none.

Not sure what the plan is exactly. Eric and Josh are running the show My only serious concern was the dictionary. We knew right off the bat that they were splitting that up and immediately was worried about everyone trying to edit the same file. We ended up sending txt files to Eric which resolved that potential issue. My other concern is how they are going to split up the rest of the work. I have a feeling that Eric and Josh are going to take the brunt of the work in order to have a successful experiment.
 * Plan:
 * Concerns:

Week Ending April 16, 2013
Try to improve the score of the last 5 hours of data
 * Task:

4/13 - Generated the transcript, moved over our dictionary and the filler. Ran into no errors. I asked Eric to look at the dictionary and the filler. The dictionary had what looked to just be a bunch of junk at the beginning and the filler only had SIL in it. Will check to see how the experiment is progressing tomorrow. There is not a lot left to do but a lot more people that may want to help, so I will hold off on doing anymore for now
 * Results:

4/14 - The previous experiment ran into issues decoding and Eric sent out emails about starting a new experiment. I went to see what I could do to help, but someone decided they would do it all themselves for no apparent reason. Kind of ticked off by this

4/15 - Looked for more work in our group folder. There is an experiment that is one lower than the latest one finished, I'm not sure if it is old and I missed it or new but I'm going to leave it alone for now. Read logs

4/16 - Logs

Run new experiments on the last 5 hours None that I can think of
 * Plan:
 * Concerns:

Week Ending April 23, 2013

 * Task:


 * Results:

4/21 - Read logs


 * Plan:

Improve our experiments
 * Concerns:

I'm not sure what I could do individually to do this

Week Ending April 30, 2013
Continue running experiments. Try to figure out why are results are worse on longer experiments.
 * Task:

4/27 - Read logs
 * Results:

4/28 - Going to try to work on the errors in the logs. What I put in the group log


 * Went through log directory for 0089
 * Found all errors and warnings

1608 times. Most of the numbers would change, except the 0 and line# utt> 3734       sw4925A-ms98-a-0013  205    0    76 23 ERROR: "backward.c", line 431: final state not reached ERROR: "baum_welch.c", line 331: sw4925A-ms98-a-0013 ignored

1228 times. This error only showed up in Module 50. Only the mgau, density and component numbers would change ERROR: "gauden.c", line 1700: var (mgau= 1099, feat= 0, density=7, component=17) < 0

3 times. Module 30 WARNING: "mod_inv.c", line 257: n_top 8 > n_density 1. n_top <- 1 WARNING: "mod_inv.c", line 257: n_top 8 > n_density 2. n_top <- 2 WARNING: "mod_inv.c", line 257: n_top 8 > n_density 4. n_top <- 4

4 times. Module 50. Exactly the same each time WARNING: "accum.c", line 626: The following seno never occur in the input data

4/29 - logs

4/30 - Was going to go through the errors for Exp 24. I was told that was not necessary and the rest of the other experiments are complete. Nothing to do here.


 * Plan:


 * Concerns:

Week Ending May 7, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns: