Speech:Exps 0194

From Openitware
Jump to: navigation, search

Description

Author: David Meehan

Date: 2/24/2014

Purpose: Attempt to decode experiments 0171, 0173, 0174, 0175 and 0176 using the test data for last_5hr.

Details: It appears as though the density is not having an impact on the score. My guess is that there is simply not enough common data between the first_5hr and the last_5hr to be successful. Before I proceed I want to get an idea for what percentage they have in common. 0174 is still processing though.

The first_5hr dictionary contains 70% of the last_5hr test data dictionary words. It appears our problem is that the first_5hr data is much too small to use for accurate decoding. Colby got better results with the 10hr, although still not great. I am abandoning the remainder of these decodes.

Acoustic Model: 0171
Language Model: 0171
Density: 8

      ,-----------------------------------------------------------------.
      |                            hyp.trans                            |
      |-----------------------------------------------------------------|
      | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
      |=================================================================|
      | Sum/Avg |  437   6474 | 51.6   43.1    5.3   20.7   69.0   99.1 |
      |=================================================================|
      |  Mean   | 36.4  539.5 | 52.1   42.9    5.1   21.9   69.8   99.3 |
      |  S.D.   |  8.3  143.2 |  4.2    3.7    2.5    6.9    6.0    1.4 |
      | Median  | 32.5  546.5 | 52.2   43.1    4.1   22.0   67.8  100.0 |
      `-----------------------------------------------------------------'

Acoustic Model: 0173
Language Model: 0171
Density: 16

      ,-----------------------------------------------------------------.
      |                            hyp.trans                            |
      |-----------------------------------------------------------------|
      | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
      |=================================================================|
      | Sum/Avg |  437   6474 | 51.0   43.6    5.4   19.7   68.7   99.5 |
      |=================================================================|
      |  Mean   | 36.4  539.5 | 51.3   43.6    5.1   20.7   69.4   99.7 |
      |  S.D.   |  8.3  143.2 |  3.4    3.1    2.7    6.8    6.4    0.8 |
      | Median  | 32.5  546.5 | 50.9   44.0    4.0   21.4   68.4  100.0 |
      `-----------------------------------------------------------------'
Acoustic Model: 0174
Language Model: 0171
Density: 32
      ,-----------------------------------------------------------------.
      |                            hyp.trans                            |
      |-----------------------------------------------------------------|
      | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
      |=================================================================|
      | Sum/Avg |  437   6474 | 48.3   45.3    6.4   19.3   71.0   99.1 |
      |=================================================================|
      |  Mean   | 36.4  539.5 | 49.0   44.8    6.2   20.2   71.2   99.3 |
      |  S.D.   |  8.3  143.2 |  4.7    4.0    3.3    6.9    6.0    1.7 |
      | Median  | 32.5  546.5 | 48.3   45.8    5.5   20.9   68.9  100.0 |
      `-----------------------------------------------------------------'

Acoustic Model: 0175
Language Model: 0171
Density: 64

Acoustic Model: 0176
Language Model: 0171
Density: 128