Speech:Exps 0095

From Openitware
Jump to: navigation, search

Mini/train test on Train (0089)


Description

Author: Group B&C

Date: April 28, 2013

Purpose: We are determining if the acoustic model from experiment 0089 is superior to the one in Experiment 0024.

Details: In experiment 0024, we ran a train and decode on an hour long corpus; curiously, the score was better than what was found in a 5 hour train. In this experiment, we are testing if the models created with a 5 hour corpus will result in a better score than what was found in experiment 0024.

Results


The Experiment dictionary required the following words to be added:

ABOU AH0 B AW1 T
ACCE AE0 K S EH1
B B IY1
BER B EH1 R
BO B OW1
CHAN CH AE1 N
CHOWPHERD  CH AW1 P ER0 D
DA D AA1
DRA D R AO1
EVERYT EH1 V R IY0 T
EXC IH0 K S
EXPRE IH0 K S P R EH1
F  EH1 F
FEDERALDES F EH1 D ER0 AH0 L EH1 S
G JH IY1
GRO G R OW1
HA HH AA1
INSPIR IH0 N S P
JO  JH OW1
L EH1 L
MARYL   M EH1 R AH0 L
MAYB M EY1 B
NOTHER N AH1 DH ER0
OI OY1
P P IY1
PEO P IY1
PER P ER1
PLAT P L AE1 T
R AA1 R
S EH1 S
SE S IY1
SH SH IY1
SOLI S AA1 L AH0
SP S P
ST S T
SU S UW1
T T IY1
TH DH AH0
THI TH ER1
TR T R
V V IY1
W D AH1 B AH0 L Y UW0
WA W AA1
WH W EH1
WI W IH1
WOU W IH1
Y W AY1
YE Y EH1

The following score was created during this experiment:

                     SYSTEM SUMMARY PERCENTAGES by SPEAKER

      ,-----------------------------------------------------------------.
      |                         hyp.trans.uniq                          |
      |-----------------------------------------------------------------|
      | Sum/Avg |  549  10707 | 43.6   50.2    6.2   19.1   75.5   99.8 |
      |=================================================================|
      |  Mean   |  2.9   56.7 | 45.2   49.7    5.1   27.9   82.7   99.8 |
      |  S.D.   |  1.9   43.4 | 15.2   15.3    4.8   32.3   32.4    2.4 |
      | Median  |  3.0   46.0 | 44.4   50.0    4.3   18.2   76.9  100.0 |
      `-----------------------------------------------------------------'