TIMIT Data Preparation Issue

[Avijoy Chakma]

Hi All,

I'm new to KALDI. Currently trying to run TIMIT scripts but stuck due to the file format issue.

I downloaded the TIMIT dataset from this link - https://www.kaggle.com/mfekadu/darpa-timit-acousticphonetic-continuous-speech

Following this link - https://github.com/srvk/eesen/issues/57, I edited the  timit_data_prep.sh (using - awk '{printf("%s '$sph2pipe' -f wav %s \n", $1, $2);}' < ${x}_sph.scp > ${x}_wav.scp) and removed the piping as suggested in the link. The output is attached as Figure-1.

Following this link - https://groups.google.com/g/kaldi-help/c/q6ydwWJiNcA/m/0P8n70LxCQAJ, I also tried without using - awk '{printf("%s %s \n", $1, $2);}' < ${x}_sph.scp > ${x}_wav.scp.  Output is attached as Figure-2.

In my train_sph.scp, the initial two lines are - 

FAEM0_SI1392 /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV

FAEM0_SI1392.WAV /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV.wav

Is there anything I can do at this point? I'm stuck here for two days.  Just to let you know that I can successfully execute the default "yesno" example in the "egs" folder. I greatly appreciate any help. Thank you.

Avijoy  

[nshm...@gmail.com]

> In my train_sph.scp, the initial two lines are - 

FAEM0_SI1392 /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV

FAEM0_SI1392.WAV /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV.wav

You have data preparation issue earlier here since you mix both NIST SPH files with WAV extension and PCM WAV files with WAV.wav extension. You need to pick either first or second. For first you need to have lines like this in wav.scp:

FAEM0_SI1392 sph2pipe -f wav /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV - |

for the second wav.scp need to look like this:

FAEM0_SI1392 /home/avijoychakma/Downloads/Acoustic/Github/Kaldi/kaldi/egs/timit/s5/data/data/TRAIN/DR2/FAEM0/SI1392.WAV.wav

Note that the first column (utterance id) doesn't have .WAV.

If you are not certain how to prepare such data with awk, you can use the scripting language you know to prepare the data, for example, Python.

[Avijoy Chakma]

Thank you so much for pointing that out. I remove the .wav files and headed with the NIST SPH files and the program executed successfully. 

Avijoy

--
Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/ee421e39-2fba-41cb-b8a5-e82e2e320f4en%40googlegroups.com.