Speech:Exps 0283 014

Description
Authors: Jon Shallow

Date: 3/24/16

Purpose: Test the newly generated utt files after adding --bits and --encoding options in the sox command in the genUttAudio script. Testing on newly created 150hr corpus.

Details: Ran into this, never seen it before. The hyp.trans looks fine though. [jrs1036@caesar DECODE]$ /mnt/main/scripts/user/parseDecode.pl decode.log ../etc/hyp.trans rm: remove write-protected regular empty file `../etc/hyp.trans'? yes
 * Train configuration
 * All default values
 * Started train at 9:33 PM on 3/24/2016
 * Train ended at 12:43 AM on 3/25/2016
 * Decode values
 * 1000 files @ 1000 senones
 * Decode started at 9:03 AM on 3/25/2016
 * Decode ended at 10:00 AM on 3/25/2016
 * Scoring
 * UPDATE
 * Ran another decode on /mnt/main/corpus/switchboard/145hr/test/trans/train.trans
 * Decode started at 9:48 AM on 3/26/2016
 * Decode ended at 12:58 PM on 3/26/2016

Results:

SYSTEM SUMMARY PERCENTAGES by SPEAKER ,-.     |                            hyp.trans                            | |-|     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |-+-+-|     | sw2001b |   14     96 | 59.4   21.9   18.8    9.4   50.0   64.3 | |-+-+-|     | sw2001a |   15    175 | 48.6   40.0   11.4    2.3   53.7   60.0 | |-+-+-|     | sw2005a |   23    447 | 49.2   28.6   22.1    2.5   53.2   91.3 | |-+-+-|     | sw2005b |   38    362 | 34.3   41.7   24.0   15.2   80.9  100.0 | |-+-+-|     | sw2006b |   26    682 | 30.5   39.1   30.4    1.6   71.1  100.0 | |-+-+-|     | sw2006a |   14    305 | 38.7   36.4   24.9    5.6   66.9   92.9 | |-+-+-|     | sw2007b |   40    603 | 44.4   33.0   22.6    2.5   58.0   92.5 | |-+-+-|     | sw2007a |   37    443 | 49.9   23.5   26.6    1.6   51.7   70.3 | |-+-+-|     | sw2008b |   17    331 | 49.8   32.0   18.1    1.8   52.0   76.5 | |-+-+-|     | sw2008a |   11    166 | 42.2   40.4   17.5   15.7   73.5  100.0 | |-+-+-|     | sw2009a |   15    221 | 36.7   38.5   24.9    7.7   71.0  100.0 | |-+-+-|     | sw2009b |   23    309 | 48.5   27.8   23.6    0.6   52.1   65.2 | |-+-+-|     | sw2010b |    9    196 | 32.7   35.2   32.1    2.0   69.4  100.0 | |-+-+-|     | sw2010a |   25    352 | 41.8   38.1   20.2    3.7   61.9   88.0 | |-+-+-|     | sw2012a |   27    495 | 23.2   31.9   44.8    1.8   78.6   96.3 | |-+-+-|     | sw2012b |   24    610 | 47.7   29.2   23.1    1.3   53.6   87.5 | |-+-+-|     | sw2013b |   51    865 | 28.3   38.6   33.1    3.0   74.7   94.1 | |-+-+-|     | sw2013a |   14    210 | 31.0   45.7   23.3    3.8   72.9  100.0 | |-+-+-|     | sw2014b |   12    267 | 35.6   44.2   20.2    4.9   69.3  100.0 | |-+-+-|     | sw2014a |   12    169 | 37.9   37.9   24.3    4.7   66.9   91.7 | |-+-+-|     | sw2015a |   14    330 | 33.3   40.0   26.7    2.1   68.8  100.0 | |-+-+-|     | sw2015b |   18    383 | 42.3   29.8   27.9    2.1   59.8   88.9 | |-+-+-|     | sw2017a |   22    203 | 41.4   38.4   20.2   13.3   71.9  100.0 | |-+-+-|     | sw2017b |   21    439 | 49.2   31.4   19.4    2.3   53.1   76.2 | |-+-+-|     | sw2018b |   26    486 | 39.7   35.2   25.1    1.9   62.1   84.6 | |-+-+-|     | sw2018a |   27    320 | 37.5   42.8   19.7   12.5   75.0   96.3 | |-+-+-|     | sw2019b |   31    504 | 48.0   30.0   22.0    1.8   53.8   71.0 | |-+-+-|     | sw2019a |   27    380 | 34.7   41.8   23.4    6.8   72.1  100.0 | |-+-+-|     | sw2020a |   33    422 | 40.0   40.0   19.9   13.3   73.2  100.0 | |-+-+-|     | sw2020b |   27    641 | 48.0   29.3   22.6    4.8   56.8  100.0 | |-+-+-|     | sw2022a |   17    325 | 31.7   38.5   29.8    2.5   70.8   94.1 | |-+-+-|     | sw2022b |   20    279 | 33.7   32.3   34.1    4.7   71.0   95.0 | |-+-+-|     | sw2023a |   39    669 | 30.6   44.1   25.3    3.1   72.5  100.0 | |-+-+-|     | sw2023b |   34    303 | 39.6   37.0   23.4    7.9   68.3   94.1 | |-+-+-|     | sw2024b |   24    343 | 39.1   32.7   28.3    1.5   62.4   83.3 | |-+-+-|     | sw2024a |   14    158 | 53.8   31.0   15.2   11.4   57.6   92.9 | |-+-+-|     | sw2025a |   18    464 | 22.6   39.2   38.1    1.9   79.3  100.0 | |-+-+-|     | sw2025b |   20    391 | 40.4   33.2   26.3    7.2   66.8  100.0 | |-+-+-|     | sw2027b |   24    411 | 37.0   29.2   33.8    6.1   69.1  100.0 | |-+-+-|     | sw2027a |   25    611 | 33.4   30.4   36.2    2.3   68.9   92.0 | |-+-+-|     | sw2028a |   38    361 | 54.0   24.7   21.3    5.8   51.8   81.6 | |-+-+-|     | sw2028b |   34    340 | 46.2   31.8   22.1    0.3   54.1   85.3 | |=================================================================|     | Sum/Avg | 1000  16067 | 39.2   34.7   26.0    4.2   65.0   90.5 | |=================================================================|     |  Mean   | 23.8  382.5 | 40.2   34.9   24.9    4.9   64.8   90.6 | | S.D.   |  9.5  166.2 |  8.4    5.8    6.4    4.2    9.0   11.3 | | Median | 23.5  356.5 | 39.7   35.2   23.5    3.1   67.6   94.1 | `-'

So, this is potentially discouraging. Maybe tweaking the train configuration file to optimal values will make a significant impact.

Results of the second decode: SYSTEM SUMMARY PERCENTAGES by SPEAKER ,---.    |                            hyp.trans                              | |---|    | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err    S.Err | |-+-+---|    |===================================================================|     | Sum/Avg | 3898  57265 | 42.0   34.2   23.8    3.6    61.6    89.1 | |===================================================================|    |  Mean   |  1.3   18.6 | 53.6   28.7   17.7    8.7    55.1    89.5 | | S.D.   |  0.5   15.9 | 25.4   18.3   15.6   31.1    35.2    28.7 | | Median |  1.0   14.0 | 50.0   31.6   17.5    0.0    59.8   100.0 | `---'