Speech:Exps 0303 006

Description
Author: Wesley Couturier

Date: 1-29-2018

Purpose: This is the first successful run of a train, language model, and decode performed by Wesley Couturier. It is used primarily as a learning requirement to represent understanding of the process in general. All students must undergo a similar sub-experiment.

Details: This process, although simplified in its current documentation, can still be further refined to both understand the process, and make this learning easier. The source for the directions I will be using are under our current processes of building a model: Speech:Models, of which I am taking references of all three steps.

The first step is to run a train. This involves first connecting to Caesar, going into the Experiments directory and traverse down the proper experiment and sub-experiment that you make / have already made manually. We then run the pearl script, makeTrain.pl switchboard 30hr/train. This makes the directory structure, tells which directory to pull information from, in this case it's located in switchboard. It also creates a soft link pointing to the WAV files of various sentences totalling 30 hours, and a few small tasks.

The next step is to create the language model for the model builder to use. It is used to determine the probability of the word(s) being used. From the root directory of our sub-experiment, we create a new directory call, enter it, then copy over the correct transcript from corpus, of the same length in time as our train, into a new file we will call. When looking in the file, we see a reference to the WAV file used, as well as the sentence as it is translated. Usually, the sentence contains unexplained errors and grammar, such as the usage of brackets. To clean up the data, we parse it by running the command parseLMTrans.pl trans_unedited trans_parsed which puts the scrubbed data into a new file. We finally copy another file that creates the language model, then run it by the following commands: cp -i /mnt/main/scripts/user/lm_create.pl . Then ./lm_create.pl trans_parsed

The final step is to run the decoder and scoring. Being within the software group, I will encourage you to go to the Spring 2018 Software Group to view more in-depth details of how the decoder works.

Results: SYSTEM SUMMARY PERCENTAGES by SPEAKER

,-.     |                            hyp.trans                            | |-|     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |-+-+-|     | sw2001b |    1     16 | 87.5    6.3    6.3    0.0   12.5  100.0 | |-+-+-|     | sw2005b |    2      6 |100.0    0.0    0.0   50.0   50.0  100.0 | |-+-+-|     | sw2006b |    1      3 | 66.7   33.3    0.0  100.0  133.3  100.0 | |-+-+-|     | sw2007b |    3     44 | 54.5   34.1   11.4   13.6   59.1  100.0 | |-+-+-|     | sw2008a |    1     11 | 63.6   36.4    0.0   18.2   54.5  100.0 | |-+-+-|     | sw2009a |    1      9 | 33.3   66.7    0.0   11.1   77.8  100.0 | |-+-+-|     | sw2010b |    1     48 | 47.9   33.3   18.8    6.3   58.3  100.0 | |-+-+-|     | sw2012a |    2     29 | 65.5   31.0    3.4    6.9   41.4  100.0 | |-+-+-|     | sw2013b |    1      3 | 66.7   33.3    0.0  100.0  133.3  100.0 | |-+-+-|     | sw2013a |    1      6 |100.0    0.0    0.0   16.7   16.7  100.0 | |-+-+-|     | sw2014a |    1     14 | 50.0   42.9    7.1   14.3   64.3  100.0 | |-+-+-|     | sw2015b |    1     43 | 79.1   16.3    4.7    4.7   25.6  100.0 | |-+-+-|     | sw2017b |    1     24 | 45.8   45.8    8.3    0.0   54.2  100.0 | |-+-+-|     | sw2018a |    1     14 | 50.0   50.0    0.0   21.4   71.4  100.0 | |-+-+-|     | sw2018b |    1     26 | 57.7   23.1   19.2   15.4   57.7  100.0 | |-+-+-|     | sw2019b |    2     42 | 88.1    4.8    7.1    9.5   21.4  100.0 | |-+-+-|     | sw2020b |    2     28 | 46.4   53.6    0.0   25.0   78.6  100.0 | |-+-+-|     | sw2022a |    1      5 | 80.0   20.0    0.0   80.0  100.0  100.0 | |-+-+-|     | sw2023a |    2     23 | 65.2   30.4    4.3    8.7   43.5  100.0 | |-+-+-|     | sw2023b |    1      6 | 83.3   16.7    0.0   16.7   33.3  100.0 | |-+-+-|     | sw2024a |    1      4 | 50.0   25.0   25.0    0.0   50.0  100.0 | |-+-+-|     | sw2025a |    2     35 | 20.0   71.4    8.6    0.0   80.0  100.0 | |-+-+-|     | sw2027b |    1     14 | 35.7   57.1    7.1    0.0   64.3  100.0 | |-+-+-|     | sw2028b |    3     11 | 54.5   36.4    9.1   27.3   72.7  100.0 | |-+-+-|     | sw2032b |    2     10 | 90.0   10.0    0.0   40.0   50.0   50.0 | |-+-+-|     | sw2035b |    1     33 | 60.6   27.3   12.1    3.0   42.4  100.0 | |-+-+-|     | sw2035a |    1      3 |100.0    0.0    0.0   66.7   66.7  100.0 | |-+-+-|     | sw2036a |    2     20 | 55.0   30.0   15.0    0.0   45.0  100.0 | |-+-+-|     | sw2038a |    1     39 | 59.0   28.2   12.8    0.0   41.0  100.0 | |-+-+-|     | sw2039b |    3     22 | 81.8   18.2    0.0   13.6   31.8   66.7 | |-+-+-|     | sw2040a |    1     16 | 62.5   31.3    6.3   12.5   50.0  100.0 | |-+-+-|     | sw2040b |    1     14 | 78.6    7.1   14.3    0.0   21.4  100.0 | |-+-+-|     | sw2041a |    1     30 | 30.0   50.0   20.0    3.3   73.3  100.0 | |-+-+-|     | sw2041b |    1      6 | 50.0   50.0    0.0   16.7   66.7  100.0 | |-+-+-|     | sw2044a |    2     64 | 68.8   23.4    7.8    1.6   32.8  100.0 | |-+-+-|     | sw2045a |    1     10 | 90.0    0.0   10.0    0.0   10.0  100.0 | |-+-+-|     | sw2045b |    1     30 | 16.7   13.3   70.0    0.0   83.3  100.0 | |-+-+-|     | sw2050b |    2     59 | 52.5   32.2   15.3   10.2   57.6  100.0 | |-+-+-|     | sw2051b |    2     29 | 51.7   34.5   13.8    3.4   51.7  100.0 | |-+-+-|     | sw2051a |    1      6 | 33.3   66.7    0.0    0.0   66.7  100.0 | |-+-+-|     | sw2053a |    1      4 |100.0    0.0    0.0   25.0   25.0  100.0 | |-+-+-|     | sw2053b |    1      3 | 66.7   33.3    0.0  133.3  166.7  100.0 | |-+-+-|     | sw2054b |    3     24 | 62.5   33.3    4.2    8.3   45.8  100.0 | |-+-+-|     | sw2055a |    1     18 | 72.2   27.8    0.0   11.1   38.9  100.0 | |=================================================================|     | Sum/Avg |   63    904 | 58.7   30.4   10.8    9.6   50.9   96.8 | |=================================================================|     |  Mean   |  1.4   20.5 | 63.0   29.2    7.8   20.3   57.3   98.1 | | S.D.   |  0.7   15.7 | 21.3   18.7   11.8   30.4   31.5    9.0 | | Median |  1.0   16.0 | 62.5   30.7    5.5   10.6   52.9  100.0 | `-'

Successful Completion