Speech:Spring 2012 Aaron Jarzombek Log

From Openitware
Jump to: navigation, search


Week Ending February 6th, 2012

Task
  • Wednesday:

Note:

I found this page looking for a download for Sphinx3. It has a lot of useful information about setting up Sphinx3. http://sphinx.subwiki.com/sphinx/index.php/Main_Page.
  • Thursday:
  • Reinstall openSUSE with the KDE interface. KDE is more graphical and a little easier to use.
  • Search for a .rpm file of Sphinx 3.08 and fail for hours.
  • Brice I created this install tutorial. VB/SUSE/Sphinx Install Tutorial

Note:

I have spent a lengthy amount of time trying to get Sphinx installed in openSUSE. YAST is useless unless you know what repositories you're looking for. If anyone reading this has any useful information on how to get Sphinx installed on openSUSE leave me a comment in the comment section below. Also for those of you who are wondering which interface to use (KDE or GNOME) KDE is a little more user friendly.
  • Saturday:
  • Read others logs.
  • Still trying to get Sphinx3 working. There must be a way if it got installed on the servers...
  • Monday:
  • Read others logs.
Results
This week was a little tricky because I expected to find Sphinx 3 for SUSE, but I failed.
Plan
Discuss it with the group and Professor Jonas to try to find a solution.
Concerns
Having to use Sphinx 4 in place of 3, then having to do more work on the back end to get it to function properly.

Week Ending February 13, 2012

Task
  • Re-format Wiki page to match Prof. Jonas's example.
  • I am going to keep working on installing Sphinx3.
  • I will look into other speech tools necessary to capture and analyze information.
  • I hopefully will be able to start working on the settings for sphinx.
Results
  • Tuesday:
  • I changed the layout of my Wiki page to match the template.
  • I sent a message to the CMU IRC channel, but nobody responded.
  • I did some more digging for the Sphinx 3 .rpm, still no luck.
  • I tried some torrent sites and other search engines besides google.
  • Wednesday:
  • I dug around on the CMUSphinx SourceForge page and found some interesting pages.
  • The first of which was about tuning Sphinx3. Since we are not really in the stages of tuning Sphinx I did not spend that much time reading the page. It does say that it is an incomplete guide, so there are most likely some gaps that we will have to research. Here is a link to the page: http://cmusphinx.sourceforge.net/wiki/decodertuning
  • The next page I read up on was about tuning a language model. This page is specific to Sphinx3. From what I understand, if we train our acoustic model as described on the page it has the capability of reducing word error rate by up to 25%, also the decoder runs faster. The cons of this model are that it requires more intensive training, sometimes requiring some samples to be done multiple times. Over all it looks like it would be a little more work on the front end, but it would pay off once we get over the hump. I don't know much about training other acoustic models, so I'll keep my eyes peeled to see if I find something better. Here is the link to the acoustic model training page: http://cmusphinx.sourceforge.net/wiki/ldamllt
  • I'm not sure if we will be using any Python to train language models, but this page has information about the layout of a ShpinxTrain with directions for C and Python. Here is the page about SphinxTrain: http://cmusphinx.sourceforge.net/wiki/sphinxtrainwalkthrough
  • This last page has some details about the features of Spinx and the formatting specifications. Here is the page about layout and formatting: http://cmusphinx.sourceforge.net/wiki/codingstyle


  • Sunday:
  • I read classmates logs.
  • Also I can't log into methusalix anymore. I had to change my password because it was too simple, and I'm assuming that it did not update. If anyone remembers what we did in class to sync our passwords with our servers please let me know.
  • Monday:
  • I read classmates logs.
Plan
  • Starting next week I'm going to install Sphinx 4, because 3 is impossible to get apparently.
Concerns
  • I would really like to stick with Sphinx 3 but it's not easy to obtain. Potential discussion about putting Sphinx 4 on the servers to keep everything the same, if we cant get 3 installed on local machines.

Week Ending February 20, 2012

Task
  • Install Sphinx 3.
  • Research on decoder.
  • Meet on Friday @ 3 on Skype with Prof. Jonas and team.
  • Write up the proposal.
Results
  • Tuesday:
  • Today I got sphinx3 installed on my local machine. I now have to figure out how to use the software. It says that its installed in /usr/local, but I cannot find the install file. Once I can actually open it I can start plugging away at the decoder tools.
  • I helped Brice write up the install instructions that we developed to install sphinx3.
  • This installation guide helps install all of the additional software packages that are required by Sphinx3 for its installation.
  • Friday:
  • Today I met with Chad, Brice and Prof. Jonas on Skype.
  • We went through some useful commands, and learned how to filter out words from the CMU dict file.
  • We ran a sample Decode to see how accurate the old train was. It is not very accurate.
  • I reviewed the Training page that Prof. Jonas has fixed up and made more user friendly.
  • The page has the Unix commands in red to make it easier to run.
  • Sunday:
  • Read classmates logs.
  • Monday:
  • Read classmates logs.

Week Ending February 27, 2012

Task
  • Write proposal.
  • Understand how perl scripts work.
Results
  • Tuesday:
  • I worked with Brice on our section of the proposal.
  • I created a text document with some sample emacs commands and code from our lab in class.
  • I started to look around in some of the perl scripts to begin figuring out how they work.
  • We broke down the run_decode script in class and it gave me a good idea of how to work with the others.
  • Thursday:
  • Today I wrote a rough draft of our proposal and sent it off to Brice for addition and editing.
  • I poked around on caesar while writing the proposal to get a better understanding of what I am doing.
  • Sunday:
  • Monday:
  • Read classmates logs.

Week Ending March 5, 2012

Task
  • Modify proposal.
  • Run sample train (train2) on ceasar.
  • Run a decode off train2.
Results
  • Tuesday:
  • Today I started out looking for important scrpits. This led me nowhere.
  • After prof Jonas came over and showed Brice and myself the direction that we should be headed in we made some good headway.
  • First we had to get our train set up.
  • We created the train2 directory so we don't break the train1 directory.
  • This involved some vi editing and changing file names to train2.
  • We got the train2 to run and have a total of 8 results.
  • Friday:
  • Read classmates logs.
  • Saturday:
  • Read classmates logs.
  • Sunday:
  • Ran a decode on the sample train with Brice
  • We conference via Skype to work through it.
  • The steps on the wiki were helpful.
  • We checked on the decode.log file and got a lot of information but, we are not sure what exactly it all means. Further research is required. Prof. Jonas might be able to shed some light on the subject.
  • Discussed the format for how we will lay out the scripts that are used.
  • A table on the wiki will be used.
  • It will be a wiki table, and not an image.

Week Ending March 19, 2012

Task
  • Create a table with useful scripts and a brief description of what they do.
Results
  • Tuesday:
  • Today Brice and I got a successful decode.
  • We had to modify some of our train information to match with train2 directory.
  • I created a table in wiki format that we can

link to an official page once it is complete.

  • Friday:
  • Read classmates logs.
  • Saturday:
  • Worked on backtracking the steps Brice and I did in our sample train.
  • I realized that the wav files that we imported were a larger folder than the sample train1.
  • I'm not sure what sample train1 used, but Brice and I used all the files.
  • Sunday:
  • Read classmates logs.

Week Ending March 26, 2012

Task
  • Finish table and add it to information page for my group.
  • Study how the scripts function.
Results
  • Tuesday:
  • Brice and I looked up some more information about the scripts and added it to the table.
  • I wrote the table up in wiki.
  • After we had a solid base of information I added the table to our group page under information.
  • Friday:
  • Read classmates logs.
  • Saturday:
  • Read classmates logs.
  • Sunday:
  • Looked into the CMU Sphinx website to see if they had any documentation about what the scrips do.
  • For the amount of searching, I didn't get a lot of useful information.
  • I learned what the make feats script does.
  • I posted the information on the notes page here

Week Ending April 2, 2012

Task
  • Brice and I are going to try and run a sample train on new data.
  • We are going to use the last 100 utterances of the sample data.
Results
  • Tuesday:
  • Today Brice and I spent our time trying to copy over the last 100 utterances from the vocab file to our wavTemp directory.
  • Initially I though that tail -100 would copy the information to the directory.
  • After I found out that I was copying text instead of text files I realized that I needed to create a file to copy the text to.
  • Brice got the TA to come help us create a file with the nano command
  • We were able to copy the information over to the vocab folder in our wavTemp folder.
  • The command we used was: tail -n 100 /media/data/Switchboard/transcripts/ICSI\ Transcriptions/trans/icsi/vocab >> speechtools/SphinxTrain-1.0/train4/wavTemp/vocab


  • Thursday:
  • Read classmates logs.
  • Saturday:
  • Read classmates logs.
  • Sunday:
  • I worked on running the train 4 with some different data for a dictionary.
  • I copied the vocab file out of wavTemp, since that should only have .sph files inside of it.
  • I put the vocab file in the etc directory and renamed it to trans_unedited.txt.
  • I copied over the .sph files from the train1 directory, but they were not sufficient.
  • Brice helped me locate the correct file when we did train 2, so I sent him an email with what I've done.
  • If he has questions about the rest of the training operation I will be around to assist him tonight.
  • I'm hoping that Brice is able to get the .sph files we need, because after that it should be smooth sailing.

Week Ending April 9, 2012

Task
  • Run my new group through what Brice and I have been doing.
  • Modify notes page.
Results
  • Thursday:
  • Prepared my notes to try and make meeting with group tomorrow go smoothly.
  • Friday:
  • I spoke with Brice about what his group did.
  • I set up a meeting with TeamViewer.
  • I helped Mike h and Bethany join my meeting in TeamViewer.
  • I worked out some small bugs in my notes.
  • I walked Mike h and Bethany through their own mini trains. (train12-14)
  • Unfortunately we ran into many different glitches.
  • I was able to help them out with a few of them, but we got stuck up at the ./scrpits make_feats script.
  • I showed them how the train was supposed to run with Brice and my train2 run.
  • Sunday:
  • Looked into the location of some of the executable scripts that are called in the RunAll.pl.
  • I was unable to find any useful information.
  • Brice is going to modify our table, and were going to run through the scripts and break them down.
  • Monday:
  • Read classmates logs.

Week Ending April 16, 2012

Task
  • Add some information to the top of our information page.
  • Link the information page to the training scripts page.
  • Fill in the Blanks on the table.
  • Do all the tasks on my new group page with my new group.
Results
  • Tuesday:
  • Set up Skype meeting with my group.
  • Added the link to the training models page to the model building page.
  • I changed the gnePhones.sh to .csh and added information to the empty block.
  • Changed the model building page by adding 2 links to CMU page about building LM and AM.


  • Thursday:
  • Read classmates logs.
  • Friday:
  • Went on Skype for meeting, but my group mates weren't on.
  • I worked with Brice's group for a little.
  • Jon helped get me in the right direction for coping files.
  • I was not successful in copying filed over to Methusalix.
  • I am going to spend more time on the software page when I'm on next.
  • Sunday:
  • I attempted to copy over the files on caesar to methusalix following the steps on install page .
  • I could not get past the last step.
  • I am going to try some other trouble shooting tonight if I have time.

Week Ending April 23, 2012

Task
Results
  • Thursday:
  • Read classmates logs.
  • Friday:
  • Skyped with Sky.
  • Got hung up in the Training procedure.
  • Couldn't figure out how to get the .sph files in my sample train.
  • Sunday:
  • Added info to my group page.
  • Monday:
  • Read classmates logs.

Week Ending April 30, 2012

Task
  • Get a mini train working with the modeling group.
Results
  • Thursday:
  • Read classmates logs.
  • Saturday:
  • Read email from Aaron g.
  • Tried to get make_feats to work, got nowhere.
  • I did not change anything in our project.
  • Looked into what files might need modification.
  • Sunday:
  • Tried finding a solution for Jonathan's read only problem.
  • I believe that if he cannot change the file he can copy the code to a new file with the updated info.
  • Monday:
  • Read Classmates logs.

Week Ending May 7, 2012

Task
Results
  • Tuesday:
  • Friday:
  • Saturday:
  • Sunday: