Speech:Exps 0248

From Openitware
Jump to: navigation, search

Description

Author: Colby Johnson

Date: 3/28/14

Purpose: Generate a comparison between using large and small LM's.

Details: I will be using an acoustic model built with the mini/train corpus subset (Exp 0243). I will be decoding using one LM modeled after the mini/train corpus subset itself and another modeled after the full/train. I will be checking to determine the difference in WER and xRT(Real Time factor) NOTE: These will be run on Rome


Results: Exp 0243 - Small LM (mini/train)

xRT = 0.30

                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg |  549  10774 | 79.9    8.5   11.5    4.9   25.0   85.4 |
     |=================================================================|
     |  Mean   |  2.9   57.0 | 82.2    8.1    9.6    8.2   25.9   86.7 |
     |  S.D.   |  1.9   44.5 | 15.7    8.7   13.5   16.2   20.8   25.2 |
     | Median  |  3.0   47.0 | 85.9    6.6    5.7    4.2   22.0  100.0 |
     `-----------------------------------------------------------------'

Exp 0248 - Large LM (full/train)

xRT = 0.28

                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg |  549  10774 | 79.6    9.5   10.9    5.7   26.1   87.4 |
     |=================================================================|
     |  Mean   |  2.9   57.0 | 81.7    9.1    9.2    9.3   27.7   89.2 |
     |  S.D.   |  1.9   44.5 | 15.3    8.0   13.3   18.3   21.6   21.9 |
     | Median  |  3.0   47.0 | 85.0    7.8    4.9    4.6   23.4  100.0 |
     `-----------------------------------------------------------------'

Successful Completion

The results Prove that the LM will not always increase accuracy.