Speech:Spring 2015 Kenneth Drews Log


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

Week Ending February 3, 2015

 * Task:

2/1: Read previous semesters logs. 2/2: Read previous logs, and did light research on sphinx, sphinx trainer and speech recognition software in general. 2/3: Continued to review previous logs.


 * Results:

2/1: After missing last weeks meeting I'm attempting to a get a better understanding of what is required and expected of my group.

2/3: I'm gaining a better understanding of the project as a whole but I'm still a bit curious as to what my group's contribution will be this semester.


 * Plan:

2/1: Contact someone in my group in order to get caught up.

2/3: After going through the logs of previous semesters I've decided my first task should be to set up a VM on my personal machine running Linux. The decision to made is should I install Fedora as the project is focused to make a switch or install openSuse like the previous semester. I think I'm going to install openSuse and switch if need be when the entire project has moved to Fedora. Will begin setting up the VM sometime later this week.


 * Concerns:

2/1: Missing this weeks meeting gives me a concerns as to what my group has decided to do thus far.

2/3: I'm concern about whats next, and what's going to happen after that. I'd feel more comfortable if I had a global view of the project and what each group intends to do this semester. I plan on reading the previous logs of other groups this semester to get a more complete view of the project.

Week Ending February 10, 2015
2/6: Continued reading logs from previous semesters.
 * Task:

2/8: Install Linux on a bootable flash drive and begin setting up the environment. I purchased a new flash drive to install and run openSuse on my personal laptop. The purpose of this is to verify and test any changes to the tools our group may make this year. I've had issues before attempting to run Linux virtually with programs such as virtualbox so I've decided just to run it from a flash drive. I also decided against installing Redhat mostly because I was unable to procure a "free" copy of it. I found a version of Linux at http://www.centos.org which is suppose to be a mirror of program of Redhat but did not want run into any issues with another version of Linux. It seams counter intuitive to add more software to the pile when we're already planning on switching Linux versions. Because I can not install Redhat it seems logical just to run openSuse.

2/9: Install openSuse.

2/10: Still having issues installing Linux but through research I realize my issue. Will complete tommorow


 * Results:

2/6: Continued reading logs from previous semesters.

2/8 I'm currently in the middle of downloading openSuse with my home internet connection its about a 4 hour download. Install openSuse instead of Redhat after a searching and research the best options.

2/9 I've aborted the installation of openSuse twice now because of fatal errors. I'm assuming the image created on my flash drive is curropt or there is an issue with the .iso file I downloaded. This has led me to once again begin the download.

2/6: Continued reading logs from previous semesters.
 * Plan:

2/8: Begin looking for any installation instructions in regards to the environment needed. Document all steps for setting up Caesar.

2/9: Continue with installation. I also plan on searching through the blogs for any installation procedures for the sphinx.


 * Concerns:

2/8: None at the moment but I'm sure that will change once I get Linux running.

Week Ending February 17, 2015

 * Task:

2/13: I've been looking through the folder structure on Caesar. I've also been reading the blog instructions on how to run a train. 2/15: Sent a few emails out to request and confirm the tools group has been added to Obelix so we would be able to update software. Today I received a confirmation and tomorrow I will attempt to do so. After learning the steps to complete this task I will attempt to upgrade sphinx and to re-run the same train to observe performance changes.

2/16: Caesar is apparently down tonight. I suppose this makes it my check in night.


 * Results:

2/15: Logged into Obelix to confirm my username was added.


 * Plan:

2/15: Run a train on Obelix.
 * Concerns:

2/15 None currently.

Week Ending February 24, 2015
2/19 Research: I'm looking for academic or professional papers concerning sphinx 3 and sphinx 4.
 * Task:

2/22 Checking in.

2/23 Checking in.

2/24 Research: I've been poring through the UNH databases for more information on the differences performance differences between Sphinx 3 - 4. So far I've only come up with one paper that specifically addresses this issue but in the last couple of days I've found several addressing the benefits of Sphinx 3.0. In addition I've found several explaining how the software works which would be helpful for future semesters. I plan on posting these on the media wiki along with updating the tools section.

2/19 I've found a paper detailing a trial comparing the two versions of sphinx. The paper basically states as for performance they are the same but Sphinx 4.0 is superior in the sense that it is more portable because it is java based. I've also found another paper comparing other speech recognition tool kits. I will need to contuine reading this paper in order to determine if it is useful.
 * Results:

2/24 See above.

2/19 Continuing research and reading with the intent of updating the media wiki by the end of the week. 2/24 Update the media wiki tomorrow before and during class.
 * Plan:
 * Concerns:

2/24 None.

Week Ending March 3, 2015

 * Task:

2/26 Checking in 3/1 Checking in

3/2 Added the paper I found that discusses the differences between Sphinx 3.6 and 4.0 to the media wiki.

3/3 Converted 4 more papers with speech recognition systems or experiments using the Sphinx tools. Will upload tomorrow before class. These should be helpful with experiments and creating a baseline score for accuracy within the class.


 * Results:

3/2 The PDF I had could not be uploaded to the wiki as it is not an acceptable file extension. I had to convert and split the file into three different png's. I'm not sure if this is the most efficient method but it was all that was available to me. It can be changed easily if pdf's can in fact be embedded.

3/3 Will upload files in the morning. I've decided even though the files must be uploaded as .png's they are still worth the trouble of clicking on each individual page.


 * Plan:

3/2 Continue reading the research I've found to determine which papers will be useful and worthy of uploading to the wiki. 3/3 Will upload files in tomorrow before class.
 * Concerns:

3/3 I have two or three overviews of on how speech recognition works but they pdf's are between 10-20 pages long which would become cumbersome when accessing them on the media wiki. Will discuss with Manager Jonas when he returns from Vegas/Kansas.

Week Ending March 10, 2015

 * Task:

3/5 Checking in.


 * Results:


 * Plan:


 * Concerns:

Week Ending March 24, 2015
3/18/15: Checking In.
 * Task:

3/20/15: Attempted to log into Caesar today in order to run a train. Apparently it is down or I have some other issue because I could not make an SSH connection.

3/24/15: Attempted again to log into Caesar with the same results as last week. 3/25/15: Attempted again to log into and again could not log into Caesar.


 * Results:

3/20/15: See above

3/24/15: See above

3/20/15: Will attempt to log in at a later date.
 * Plan:
 * Concerns:

Week Ending March 31, 2015
3/25/15 I plan on running my 1st five hour train in a few minutes. I admit it's a bit late in the semester to do so but catching up is my specialty. Afterwords if all goes well I will look in to making changes for performance with a 256 hour train. I will post on the super secret group page any changes made to the 256 and email my corresponding group buddy.
 * Task:

3/27/2015: Had a few permissions issues last night when trying to run a train. They were fixed by the systems group. Currently running a train. Update Later. Update (3.30.15): I've run the train but have logged in several times to run the Decode and have been unable to due a script error. I've made about 4 attempts to no avail.

3/30/15: Checking in.


 * Results:

3/25/15: None yet


 * Plan:

3/25/15: See above:
 * Concerns:

3/25/15: None yet

Week Ending April 7, 2015

 * Task:

4.2.15 Checking In. I logged into Caesar to check a script but forgot to update this blog.

4.6.15 : Running a modified train on the first 5 hours. This is more to get use to changing settings than it is to upgrade performance. I've also been reading through some of the literature I did not post in order to determine configuration settings that might be useful. I plan on running 2 trains tonight with decodes tomorrow.

4.7.15 Ran the both trains 005, 006. They had minuet changes to them. 005 failed with a script error apparently I didn't set two values equally and it basically crashed before actually running. I moved on to 006 ran the train last night and today I created the language model, ran every script in the decode except the final where I get an error. I checked the email Prof. Jonas sent concerning the changes to decode and I had followed all the appropriate script calls but I still get an error. I can't see what I'm missing or doing wrong but I will consult with the group tomorrow in order to correct this issue.


 * Results:

4.6.15 : Will give updates to group if my changes are useful.

4.7.15 One failed train, One failed decode.


 * Plan:

4.6.15: See above
 * Concerns:

Week Ending April 14, 2015

 * Task:

4.9 Checking In.

4.10 Checking in

4.14 I've been reading through the last few academic papers I've found on speech recognition. I think I'm going to post at least two more to the literature page as they give a really good overview in very simple terms how speech recognition works. I think they would be very useful for those entertaining into the project next semester. I should post them tomorrow before class. I had hesitated earlier because of their size but after re-reading them I think they would be really useful. I also finished our groups poster earlier this evening and emailed out for the groups approval.


 * Results:


 * Plan:

4.14 Post the papers tomorrow.


 * Concerns:

Week Ending April 21, 2015

 * Task:

4.20 checking in.

4.21: This week I sent out an email to every member on our team who was looking for work to be done that doesn't involve running trains. The results were pretty successful as we now have a lot more record keeping processes being completed. This should help not only for future semesters to see what our tactics were but also it will help to culminate the results we turn in at the end of the semester in regards to the competition. We are now running a weekly log of all the groups combined experiments and the results from those experiments. The group will also be keeping a list of parameter changes that were successful, unsuccessful, and indifferent. We are also cataloging our groups chain email for posterity, and another member will be creating an instruction guide on what parameters can be changed in experiments and the ranges they can be changed. These additional tasks will help the entire group with their own experiments and should make our weekly meetings more eventful as almost everyone in the group will now have something to contribute during our meetings.


 * Results:

New tasks given out to several group members.


 * Plan:


 * Concerns:

Week Ending April 28, 2015

 * Task:

4.23 Checking In

4.26 Checking In

4.27 Uploaded one of the two papers I mentioned before. The second paper was already uploaded by another student. In addition I've been looking for more information on Speech Recognition but have found very few papers that would useful. I think we exhausted the UNH library databases. Kayla and I discussed giving another student some of the extra work mentioned in last weeks blog but I'm not sure if that work was completed. As a group we're pretty well prepared to present our experiments as everyone has been working on their individual parts for the results. I'm just waiting for the final week to put all of the information together. I've really been striving to keep very good records of our groups overall work this semester. This should be very helpful for future semesters because we'll be leaving a complete record of what we did and things we overcame during the project. Future semesters will be weeks ahead of us in regards to the preparation for running experiments, strategies, and over all knowledge on speech recognition. As I haven't left much of a mark in technical terms I think the work we're doing and I've helped to organized is just as important.


 * Results:


 * Plan:


 * Concerns:

Week Ending May 5, 2015

 * Task:

5.4 I completed a rough draft of our group's results today. The results page is pretty conclusive because most people provided input into it. It was more a matter of taking people's portions and rewriting them to fit into a single document. I sent it out so everyone could review it and make suggestions for edits or alterations. I'm assuming tomorrow I will have to edit the document in some way after everyone has commented on it.

Our strategy is plainly laid out with specific parameter changes and the reason behind them. I also included a very rough explanation of speech recognition in order to make the parameter changes make sense.

5.5 I made a few changes to the results document today suggested by the group. Mostly minor changes but through discussion they had the same comments I did. I would have loved to be able to write another 2 pages. I left out a bunch of stuff concerning our documentation process and things we did to for the entire project. I had asked Kayla to create a page of specific parameter changes you can make to trains. She posted it last week and was going to post an instruction sheet with the same subject for decodes. We decided to wait until the end of the semester to post the decode instructions because we felt it was giving away to much to the other group. It should be posted by the end of the project though. In addition Morgan should be posting our chain email which has a lot of good information for future semesters.

I actually wrote about another page of stuff for the results document concerning these topics but had to leave them out because of space constraints. I also had to remove a lot of paragraph breaks to the document which makes it a little less clean but I felt I had cut enough out of the document already and did not want to sacrifice even more information.

Zach has a few changes he wanted to make personally in regards to some information he specifically wanted in the paper. I would talk about it but the competition isn't over.
 * Results:


 * Plan:


 * Concerns: