Speech:Exps 0199

From Openitware
Jump to: navigation, search

Description

Author:
Colby Johnson & David Meehan

Date:
27Feb2014

Purpose: The purpose of this Experiment if to show a baseline of accuracy using clean transcription data meaning no OOV words. This will hopefully prove where our inaccuracy lies. using varying densities and senone values we hope to show data curves of optimal values. This is one of 3 Experiments using different corpora, this one will use first_5hr/clean

Details: All trains are run with the following settings: Corpus -

  • first_5hr/clean

0199/d8/s3000 -

  • Density: 8
  • Senone: 3000

0199/d8/s5000 -

  • Density: 8
  • Senone: 5000

0199/d8/s7000 -

  • Density: 8
  • Senone: 7000

0199/d16/s3000 -

  • Density: 16
  • Senone: 3000

0199/d16/s5000 -

  • Density: 16
  • Senone: 5000

0199/d16/s7000 -

  • Density: 16
  • Senone: 7000

0199/d32/s3000 -

  • Density: 32
  • Senone: 3000

0199/d32/s5000 -

  • Density: 32
  • Senone: 5000

0199/d32/s7000 -

  • Density: 32
  • Senone: 7000

Results

  • All trains were successful
  • Decodes are in progress

0199/d8/s3000 - Train: 48 Min Decode: xRT = 1.51

  • Density: 8
  • Senone: 3000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 84.4   11.1    4.5   16.4   32.0   91.6 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 84.4   11.4    4.2   18.7   34.3   92.6 |
     |  S.D.   | 20.2  248.5 |  5.6    4.6    2.1    9.3   11.8    7.5 |
     | Median  | 40.0  489.5 | 85.6   10.4    3.8   17.3   32.9   95.3 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d8/s5000 - Train: 61 Min Decode: xRT = 1.55

  • Density: 8
  • Senone: 5000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 87.1    8.8    4.2   15.5   28.5   89.0 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 86.9    9.1    3.9   17.7   30.7   90.3 |
     |  S.D.   | 20.2  248.5 |  4.7    3.8    1.9    8.9   11.2    8.5 |
     | Median  | 40.0  489.5 | 88.0    8.6    3.5   15.5   29.6   92.3 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d8/s7000 - Train: 61 Min Decode: xRT = 1.73

  • Density: 8
  • Senone: 7000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 86.5    9.2    4.3   15.4   28.9   89.3 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 86.4    9.5    4.1   17.6   31.2   90.6 |
     |  S.D.   | 20.2  248.5 |  4.6    4.0    2.1    9.1   10.9    8.3 |
     | Median  | 40.0  489.5 | 87.2    8.9    3.7   15.1   29.3   92.3 |
     `-----------------------------------------------------------------'

Successful Completion

0199/d16/s3000 - Train: 85 Min Decode: xRT = 1.67

  • Density: 16
  • Senone: 3000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 90.1    6.4    3.5   13.7   23.7   85.5 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 90.0    6.7    3.3   15.8   25.8   86.8 |
     |  S.D.   | 20.2  248.5 |  4.0    3.3    1.7    8.3   10.0    9.4 |
     | Median  | 40.0  489.5 | 90.7    6.0    3.0   14.4   24.4   89.6 |
     `-----------------------------------------------------------------'

Successful Completion

0199/d16/s5000 - Train: 83 Min Decode: xRT = 1.62

  • Density: 16
  • Senone: 5000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 91.0    5.6    3.4   12.5   21.5   81.5 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 90.9    5.8    3.3   14.5   23.6   83.2 |
     |  S.D.   | 20.2  248.5 |  3.9    2.8    2.3    8.1    9.6   10.8 |
     | Median  | 40.0  489.5 | 92.1    4.8    2.7   12.0   21.8   85.4 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d16/s7000 - Train: 82 Min Decode: xRT = 1.53

  • Density: 16
  • Senone: 7000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 89.0    6.3    4.7   12.4   23.4   82.2 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 89.0    6.4    4.6   14.2   25.2   83.7 |
     |  S.D.   | 20.2  248.5 |  4.3    3.0    3.2    7.7    9.3    9.5 |
     | Median  | 40.0  489.5 | 89.8    5.5    3.7   12.3   22.6   85.4 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d32/s3000 - Train: 115 Min Decode: xRT = 1.82

  • Density: 32
  • Senone: 3000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER                     
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 93.1    3.9    3.0   10.3   17.2   74.2 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 93.0    4.1    2.9   12.1   19.2   76.3 |
     |  S.D.   | 20.2  248.5 |  3.2    2.1    2.1    7.2    8.3   10.8 |
     | Median  | 40.0  489.5 | 93.8    3.9    2.3    9.7   16.8   79.2 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d32/s5000 - Train: 103 Min Decode: xRT = 1.72

  • Density: 32
  • Senone: 5000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER                     
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 89.8    5.1    5.2    8.9   19.2   74.3 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 89.9    5.1    5.0   10.3   20.4   75.9 |
     |  S.D.   | 20.2  248.5 |  4.3    2.0    3.6    6.0    7.3    9.9 |
     | Median  | 40.0  489.5 | 90.6    4.9    4.2    8.5   18.5   76.6 |
     `-----------------------------------------------------------------'

Successful Completion


0199/d32/s7000 - Train: 102 Min Decode: xRT = 1.80

  • Density: 32
  • Senone: 7000
                    SYSTEM SUMMARY PERCENTAGES by SPEAKER                     
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 3506  43526 | 84.3    6.4    9.4    8.4   24.1   77.1 |
     |=================================================================|
     |  Mean   | 43.8  544.1 | 84.3    6.4    9.3    9.5   25.1   78.6 |
     |  S.D.   | 20.2  248.5 |  6.4    2.1    6.0    5.4    7.6    9.6 |
     | Median  | 40.0  489.5 | 85.7    6.0    7.7    7.7   25.3   79.2 |
     `-----------------------------------------------------------------'

Successful Completion