Speech:Exps 0103

From Openitware
Jump to: navigation, search

Forced min. Iteration: Test on Train 0101


Description

Author: Eric Beikman

Date: 6/15/2013

Purpose: To run a test on train 0101 to determine if the changed parameters have an effect on the created model.

Details: For this Experiment, we utilized the 30 minute last_5hr/test corpus to test train 0101. We bumped the minimum amount of iterations from '1' to '4' to determine if this would have a beneficial effect on the resulting model.

To enable us to compare to our previous "Best" scoring experiment for this corpus (Experiment 0090), this experiment utilizes the language model, dictionaries, and phone list from that particular experiment.

Results

The following score was created during this experiment:

      ,-----------------------------------------------------------------.
      |                            hyp.trans                            |
      |-----------------------------------------------------------------|
      | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
      |-----------------------------------------------------------------|
      | Sum/Avg |  437   6474 | 15.6    6.4   77.9    0.5   84.8   95.0 |
      |=================================================================|
      |  Mean   | 36.4  539.5 | 16.5    6.3   77.2    0.5   84.0   94.9 |
      |  S.D.   |  8.3  143.2 |  4.5    3.0    5.4    0.5    4.3    7.0 |
      | Median  | 32.5  546.5 | 17.3    6.3   75.0    0.5   83.1  100.0 |
      `-----------------------------------------------------------------'

When compared to Experiment 0090, the score is abysmal at best. The word error rate has more than doubled as a direct result of minimum amount of iterations being set to 4! We can safely assume that it is best to allow Sphinx to determine the amount of Baum-Welch iterations necessary.