Speech:Exps 0166

From Openitware
Jump to: navigation, search

Description

Author: Colby Chenard

Date: 17Feb2014

Purpose: The goal of this experiment was to test out different densities values on the 10hr train to optimize WER. This experiment was #3 of six that were done simultaneously(0162,0164,0166,0168,0170,0182) each has the same senone value
but we increased the densities each train, so 8,16,64. I skipped 32 because that was already done by Colby J in previous experiments.

Details: Colby J and I hypothesized that we may get better results as we increase the densities. The downside to this is that the higher the density the longer it takes to decode. I hope that with these experiments that we can get under 30% error rate.

Corpus/Switchboard:

  • 10hr train

Sphinx_train.cfg:

  • Semone value: 5000
  • Density: 64

Dictionary:

  • first_5hr_train_full "Master Dictionary"
    • /mnt/main/corpus/dist/custom/first_5hr_train_full.dic

GenTrans:

  • genTrans5.pl


Results The train ran in: 6 Hours 53 Min Decode ran in: 104135 Sec (28.89 Hours)

                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 8860  134108| 93.3    3.9    2.8   10.0   16.7   78.7 |
     |=================================================================|
     |  Mean   | 43.9  663.9 | 93.2    4.1    2.7   10.9   17.7   79.6 |
     |  S.D.   | 19.4  286.0 |  3.1    2.0    2.3    6.0    7.6   11.1 |
     | Median  | 40.0  595.5 | 93.9    3.6    2.1    9.8   16.4   81.3 |
     `-----------------------------------------------------------------'