Speech:Spring 2013 Josh MacPherson Log


 * Home
 * Semesters
 * Spring 2013
 * Proposal
 * Report

Week Ending February 5th, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending February 12, 2013

 * Task:

Feb 4: Group meeting over google + Feb 10: Read logs Feb 11: Read logs


 * Results:

Feb 4: Set up and successfully ran a tiny train.


 * Plan:


 * Concerns:

Week Ending February 19, 2013

 * Task:

Study the train assembly and analysis process and determine the best way to further automate the process to increase efficiency and productivity.

Read logs.


 * Results:

The first 4 scripts can be subsumed into one script that will assemble and properly configure a new training experiment. The last 4 scripts will compose another script which will run the train and handle the decoding and analysis of the experiment output.


 * Plan:

Creation of Two "super-scripts" that will call the existing series of Perl scripts to properly build and then run the train is desirable. The goal is to reduce the current process which consists of 8 scripts and numerous intermediate steps, to just a "build_train.pl" script and a "run_train.pl" script.This will reduce the set-up time significantly and make the modeling process easier to use for the other groups.


 * Concerns:

Week Ending February 26, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 5, 2013

 * Task:


 * Results:


 * Plan:


 * Concerns:

Week Ending March 12, 2013

 * Task:


 * Results:

The existing experiment set-up page is some what incomplete, following all instructions word for word, will result in a train to run, however there will be problems during decoding and scoring, so it is important to create a more complete instruction set.
 * Plan:
 * Concerns:

Week Ending March 26, 2013
Continue to run trains, decode and score experiments.
 * Task:

Create a new wiki page with more detailed/refined instructions on the modeling process.

3/20: Run and decoded 0027. Made initial notes for revised wiki page.
 * Results:

3/22: Read Logs. Held online meeting with my group, demonstrated experiment setup through to running a train (Exp# 0028). Made improvements to the modeling documentation with Eric.

3/25: Read Logs

3/26: The updates/revisions to the modeling documentation are now mostly finished, the documentation is more thorough and readability is much improved.

The modeling group will lead there respective sub-groups through the decode and scoring process after the status meeting on wednesday.
 * Plan:


 * Concerns:

Week Ending April 2, 2013

 * Task:

Troubleshoot issues that some people have been having with running a train. Get everyone ready to decode and score. Get every group C member to contribute to the group log report.


 * Results:

3/30: Read Logs.

4/1: Read Logs.


 * Plan:


 * Concerns:

Week Ending April 9, 2013

 * Task:

Train with a 5-hour corpus segment to develop an improved acoustic model. Get all members of Group BC working together on the 5-hour project. Eric and I need to create documentation and assign responsibilities for our group members using shared google docs. Remind group C members to contribute to the group log.

Set-up a shared Google folder for group BC. All group members participated in dictionary updates. Successfully completed a test on train for the 5-hour experiment. Unfortunately problems with the transcript have caused a greater word error rate than expected.
 * Results:

Read logs.

Updated group assignments.


 * Plan:


 * Concerns:

Need to work with the data group to ensure the 5-hour corpus segment is ready for experimentation.

Week Ending April 16, 2013

 * Task:

Debug the transcript/data issues that effected our 5-hour test on train. The whole BC group will again be given some assignment to work on, with the goal of improving acoustic model thus decreasing our word error rate.

Get the BC group to contribute to the BC group log.


 * Results:


 * Plan:


 * Concerns:

Week Ending April 23, 2013

 * Task:

Improve the word error rate of the five hour experiment, problems are stemming from the transcripts primarily. Have been committing to/reading the group log.

Considerable "log type' data has been recorded in our shared BC Group folder, Eric and I will make an effort to incorporate this information into the wiki Group Log.
 * Results:


 * Plan:


 * Concerns:

Week Ending April 30, 2013

 * Task:

Examine and catalog the errors/warnings that have been frequently plaguing our experiments. Determine the impact of the modified transcripts on the word-error rate by running our previous best results experiments with the new transcripts and resulting modals. Read some logs. Write some logs.

4/27: read logs
 * Results:

4/29: read logs Group BC has parsed the experiment logs and compiled a list of the most frequent errors. The AM of experiment 0024 was tested against the 5hour experiment.


 * Plan:


 * Concerns:

Week Ending May 7, 2013

 * Task:

Continue to run new experiments with the BC group. New experiment with genTrans6.


 * Results:

5/2: Read Logs

5/5: Read logs

5/7: All planed experiments have been completed. Contributed to group log.


 * Plan:


 * Concerns: