nbravos_version from https://github.com/nbravo/evaluate with slight modifications
For original README, see ORIGINAL_README
-
Make sure you run 'source path.sh' to set environment variables. Depending on how files have changed, this may be outdated so you may have to find where the scripts are.
-
Run the decoder using ./decode_and evaluate.sh 0 conf/gale.conf or ./decode_and evaluate.sh 0 conf/nutrition.conf
-
Follow below if you would like to change the wav files to decode
#--- Using conf/gale.conf ---
2.1a. Specify files in /data/sls/scratch/nbravo/kaldi/egs/gale_arabic.complete/s5/data/test/split30/nbravo/spk2utt THe numbers at the end (_1006200_1013930) represent time stamps
2.2a. Specify files in /data/sls/scratch/nbravo/kaldi/egs/gale_arabic.complete/s5/data/test/split30/nbravo/wav.scp This uses file paths in the directory /data/sls/scratch/amali/data/GALE/LDC2013S02/gale_p2_arb_bc_speech_p1_d1/data/. So if you want to add more files, use them from this path
2.3a. Specify /data/sls/scratch/nbravo/kaldi/egs/gale_arabic.complete/s5/data/test/split30/nbravo/segments. I am unsure of what the numbers are supposed to mean, particularly the _1006200_1013930 at the end of the filenames.
Note: Yes, this is nbravo's directory which may change at any point in time. And yes you must specify individual files unless you write a script to do batches.
#--- Using conf/nutrition.conf ---
Go to conf/nutrition.conf and change
2.1b. SPK2UTT path: /data/sls/scratch/pricem/kaldi/nutrition/data/test_amt1/split10/ vary this number from 1-10 /spk2utt
2.2b. WAV path: /data/sls/scratch/pricem/kaldi/nutrition/data/test_amt1/split10/ put the same number /wav.scp
Note: This does batches of wav files.
#----------------------------------
- Look at the step 1.
Notes: Haven't figured out how to make ./recognize work. But it is just a running lots of ./decode_and_evaluate. All credit to nbravo.