Modifications to $PATH if $KALDI_ROOT is not set

[vesis84]

Modifications to $PATH if $KALDI_ROOT is not set

[KarelVesely84]

Hi Quentin,
yes, getting kaldi tools to $PATH is necessary, so that they can be used in the 'pipeline filenames':
kaldi_io.read_mat('sum-matrix ... | scale-matrix ... - |')

And, yes, i could eventually put it into a try block, and show an annoying warning if Kaldi is not found...
So you don't use kaldi at all, you are just importing the its data formats in another project?
Am i correct?
K.

[KarelVesely84]

In what way was it failing to you? Do you use Linux, or other OS?
K.

[KarelVesely84]

I modified it a bit, please let me know if it solves you problem:
d6f9d9f

[qmeeus]

So you don't use kaldi at all, you are just importing the its data formats in another project?
Am i correct?

I have already computed all the features I needed with Kaldi scripts, I am indeed just using your code for loading the generated files

In what way was it failing to you? Do you use Linux, or other OS?

This section in particular is problematic:

if not 'KALDI_ROOT' in os.environ:
    # Default! To change run python with 'export KALDI_ROOT=/some_dir python'
    os.environ['KALDI_ROOT']='/mnt/matylda5/iveselyk/Tools/kaldi-trunk'

I had some problems with the encodings and was trying to traceback what was going wrong. At some point, I echo'ed my environment and saw a lot of weird things had been added to my $PATH, including a folder that did not exist in my filesystem... Turns out the code above was the reason.

I think that instead of adding this to the environment variable, which will only work for you, you should raise an error. In fact, setting $KALDI_ROOT and having Kaldi bins in the $PATH is not always necessary to use most functions in your script, so it's probably best to check and raise when Kaldi is necessary and ignore otherwise.

[KarelVesely84]

i don't understand, how extending $PATH can cause problems with encodings... even if there is a path added that does not exist, it should not cause any failures IMHO...

and having the $PATH extended only sometimes would not be user friendly... having kaldi binaries on $PATH is an important and very handy/powerful feature...

[qmeeus]

It does not cause problems with the encoding

[erip]

I'll add that it's a bit weird to include a hardcoded path that will almost certainly not exist for anyone else. If $KALDI_ROOT is really required but it's not set, it's better to throw an error than to silently set it to a path that won't necessarily exist.

[KarelVesely84]

Well, it depends. For some use-cases running kaldi binaries in pipe commands is not necessary.
In this case setting $KALDI_ROOT is not necessary. Only in the "advanced" cases with pipes it becomes necessary.
So both "modes" make sense, depending on the circumstances...
K.