Speech:Exps 0259 011

Description
Author: Justin Thibeault Purpose: Establish adaptation procedures. Details: Marcel followed the adaptation procedure using the full NOAA corpus and the CMU adaptation instructions in experiment 0257/023. Here I am using them on the split NOAA corpus.

First I ran the Baum Welch executable. It created 3 files - gauden_counts, mixw_counts, tmat_counts.

Used mllr_solve to create a file mllr_matrix. Added the mllr switch to my decode, but it seg faulted.

From what I can find in the documentation, mllr adaptation only works in pocket sphinx, so I adapted the model using just the BW adaptation and ran a new decode with excellent results.

Results
SYSTEM SUMMARY PERCENTAGES by SPEAKER

,-.     |                       results/011.hyp.txt                       | |-|     | SPKR   | # Snt  # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |+--+-|     | noa    |   58    2368 | 87.6    8.2    4.2    3.3   15.8   70.7 | |=================================================================|     | Sum/Avg|   58    2368 | 87.6    8.2    4.2    3.3   15.8   70.7 | |=================================================================|     |  Mean  | 58.0  2368.0 | 87.6    8.2    4.2    3.3   15.8   70.7 | | S.D.  |  0.0    0.0  |  0.0    0.0    0.0    0.0    0.0    0.0 | | Median | 58.0 2368.0 | 87.6    8.2    4.2    3.3   15.8   70.7 | `-'

And results on test: SYSTEM SUMMARY PERCENTAGES by SPEAKER

,-.     |                           011.hyp.txt                           | |-|     | SPKR   | # Snt  # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |+--+-|     | noa    |   57    2430 | 86.4    8.1    5.5    3.1   16.7   78.9 | |=================================================================|     | Sum/Avg|   57    2430 | 86.4    8.1    5.5    3.1   16.7   78.9 | |=================================================================|     |  Mean  | 57.0  2430.0 | 86.4    8.1    5.5    3.1   16.7   78.9 | | S.D.  |  0.0    0.0  |  0.0    0.0    0.0    0.0    0.0    0.0 | | Median | 57.0 2430.0 | 86.4    8.1    5.5    3.1   16.7   78.9 | `-'