Re: [Kaldi-developers] Poor man's Kaldi recipe

[Vassil P. <vas...@gm...>]

Please, consider my comments below.

On Wed, Feb 22, 2012 at 5:19 PM, Daniel Povey <dp...@gm...> wrote:
> I think for something like this it's not necessary to create a
> branch-- you can just get it working on your local copy, and check it
> in when it's done.  Since the stuff you're doing wouldn't break any
> functionality in anything else, I don't think we need to test it too
> thoroughly before adding it.

True. I was just wary of commiting directly in trunk, but will try to
make sure that not too many commits will be necessary(hopefully just
one) to get it right.

> RE the binary that converts CMU's
> feature files into a table-- do you have source for this?  Is it
> written in the Kaldi style?

You can see the source at:
https://github.com/vdp/kaldi-rm1-mod/blob/master/cxx/sphinx/pack-sphinx-feats.cc
. This was the first C++ Kaldi tool, that I wrote and I wasn't
particularly concerned about making it comformant in style at the
time, but as far as I could say it is mostly Kaldi-style.
As you can see it is basically a very trivial Holder class, and Sphinx
feature format is very simple itself(which in this case is not a good
thing because just the number of vectors are recorded in the header
but nothing else, not even their dimensionality). If you have
recommendations I will try to unify the style more and I could
probably think about adding some kind of automatic endianess detection
instead of hardcoding it as template parameters. Also perhaps the name
can be changed to, for example, "copy-sphinx-feats".

> We could just include it in the tools,
> e.g. in featbin/

Yes, that's where I placed it during my experiments, but frankly I
wonder if it should be stored along with the core tools, because most
likely the majority of the Kaldi users won't use it.

> BTW, the Apache copyright notice should have just your name-- it's
> whoever wrote the tool.  Hopefully you don't work for a company that
> claims ownership of whatever you do.

No, that's not an issue in my case - just my name will be in the
header. Is the notice of the file above OK?

Regards,
Vassil