Speech:Exps 0303 049

Description
Author: Daniel Beitel

Date: 2-24-2018

Purpose: This is an unseen data sub-experiment making sure there are no errors in reference to LDA experiment bugs in 032/046/047

Details:

- The purpose of me trying to run a train/test on unseen data is to rule out that making them is not the issue that modeling is running into when trying to run an LDA. I have successful run this and gotten a result which means that most likely the documentation in running the LDA. Performed on drone server Asterix.

Steps
1. addExp.pl -s on caesar then cd into /mnt/main/Exp/0303 and create matching directory (mkdir) and cd into it    2. makeTrain.pl switchboard 5h/train 3. genFeats.pl -t 4. run "top" nohup scripts_pl/RunAll.pl & 5. mkdir LM and cd into it    6. cp -i /mnt/main/corpus/switchboard/5hr/train/trans/train.trans trans_unedited 7. parseLMTrans.pl trans_unedited trans_parsed 8. cp -i /mnt/main/scripts/user/lm_create.pl. 9. ./lm_create.pl trans_parsed 10. addExp.pl -s and create a second sub-experiment then cd into /mnt/main/Exp/0303 and create matching directory (mkdir) 11. makeTest.pl -t switchboard/5hr 0303/(train directory) 0303/(new directory) 12. genFeats.pl -d 13. cd into the etc directory 14. nohup run_decode.pl 0303/(first directory made) 0303/(second directory made) 1000 & 15. parseDecode.pl decode.log hyp.trans 16.cd into first experiment etc folder 17. cp (first_directory_#)_train.trans /mnt/main/Exp/0303/(second directory #)/etc 18. cd in /mnt/main/Exp/0303/second directory/etc 19. run sclite -r (first experiment directory #)_train.trans -h hyp.trans -i swb >> scoring.log

Results:

,-.     |                            hyp.trans                            | |-|     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err | | Sum/Avg |  65   1054 | 69.4   23.0    7.6    6.6   37.2   89.2 | |=================================================================|     |  Mean   |  1.3   21.1 | 72.8   21.1    6.1   13.1   40.3   87.0 | | S.D.   |  0.5   19.3 | 21.7   19.0    9.7   21.5   26.2   33.2 | | Median |  1.0   15.5 | 72.7   20.0    1.1    3.7   37.8  100.0 | `-'