Speech:Exps 0257 001

<< 0257 Marcel NOAA exp

After trying to execute simple scrips like copy_setup.pl without any luck I decided to copy experiment 0256/001. This is first_5hr corpus train. My goal was to use this training data to decode NOAA data.

I generated the set of acoustic model feature files from these NOAA WAV audio recordings.

sphinx_fe -argfile /mnt/main/Exp/0257/001/etc/feat.params -samprate 16000 -c \ /mnt/main/Exp/0257/CorpusNOAA16/etc/CorpusNOAA16.fileids -di. -do. -ei wav -eo mfc -mswav yes

Ran the decode manually without any scripts.

/usr/local/bin/sphinx3_decode \ -hmm /mnt/main/Exp/0257/001/model_parameters/001.cd_cont_1000 \ -dict /mnt/main/Exp/0257/001/etc/001.dic \ -fdict /mnt/main/Exp/0257/001/etc/001.filler \ -lm /mnt/main/Exp/0257/LM/tmp.lm.DMP \ -ctl /mnt/main/Exp/0257/CorpusNOAA16/etc/CorpusNOAA16.fileids \ -cepdir /mnt/main/Exp/0257/CorpusNOAA16/wav \ -cepext .mfc \ -hyp /mnt/main/Exp/0257/result/firstNOAA.txt

! The LM folder from /mnt/main/Exp/0257/ was moved to  /mnt/main/Exp/0257/003/

After running the decode I got the following results:

SYSTEM SUMMARY PERCENTAGES by SPEAKER

,-.     |                          firstNOAA.txt                          | |-|     | SPKR   | # Snt  # Wrd | Corr    Sub    Del    Ins    Err  S.Err | |+--+-|     | noa    |  185    6320 | 36.6   33.3   30.1    2.0   65.4  100.0 | |=================================================================|     | Sum/Avg|  185    6320 | 36.6   33.3   30.1    2.0   65.4  100.0 | |=================================================================|     |  Mean  |185.0  6320.0 | 36.6   33.3   30.1    2.0   65.4  100.0 | | S.D.  |  0.0    0.0  |  0.0    0.0    0.0    0.0    0.0    0.0 | | Median |185.0 6320.0 | 36.6   33.3   30.1    2.0   65.4  100.0 | `-'