Speech:Exps 0080

Description
Author: Group BC

Date: 4/9

Purpose: To see how a phone set without stress indicators affects the word accuracy rate.

Details:

The goal for this experiment was to determine the word accuracy rate of an acoustic and language models without stress indicators. This experiment is inconclusive as the decoder fails for all but two audio files. The correlating acoustic model creation experiment is at Experiment 0082.

Originally this decode was to use the acoustic model generated in experiment 0074; however, as the two experiments differed in what phones they utilized, the decoder would not proceed. Rather than scrap this experiment, we continued after creating an acoustic model created using the same phone set as this experiment.

Results

The experiment utilized the same set of words as experiment 0075; however, all numeric stress indicators were stripped using sed.

The results are inconclusive as the decoder failed to initialize for all but two audio files.

The following score was created during this experiment: SYSTEM SUMMARY PERCENTAGES by SPEAKER

,-.     |                            hyp.trans                            | |-|     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |-+-+-|     | sw4809a |    6     73 | 72.6   17.8    9.6    4.1   31.5   66.7 | |-+-+-|     | sw4809b |    2     10 | 40.0   50.0   10.0   10.0   70.0  100.0 | |=================================================================|     | Sum/Avg |    8     83 | 68.7   21.7    9.6    4.8   36.1   75.0 | |=================================================================|     |  Mean   |  4.0   41.5 | 56.3   33.9    9.8    7.1   50.8   83.3 | | S.D.   |  2.8   44.5 | 23.1   22.8    0.3    4.2   27.2   23.6 | | Median |  4.0   41.5 | 56.3   33.9    9.8    7.1   50.8   83.3 | `-'