[CLAM] [Fwd: You are invited to take part in The Google Summer of Code(tm) 2008]

Georg Holzmann grh at mur.at
Tue Mar 18 14:08:04 PDT 2008


> speex support is not as prioritary as flac but it would be interesting as 
> well. Then the aubio part is pretty interesting. I would recomend you to get 
> into CLAM basics in order you to understand the challenges of having as 
> output a segmentation and propose an idea on how to address it.

OK, I see ...

> Having experience with python, numpy and swig is a very valuable point. We 
> were more interested in SIP because it seems simpler (and because pyqt uses 
> it) but we are open to change our mind provided some good arguments.

Yeah, this depends what one wants to do. I also evaluated boost python, 
SWIG and SIP for my library and it turned out that when one wants to 
integrate into numpy/scipy, SWIG was much easier to use (it is used by 
many other libs, there is also an automatic doxygen to python docstring 
conversion and etc. - but maybe this is also possible with SIP) ...
However, maybe for CLAM SIP should be also considered because of QT, so 
this should be re-evaluated in a first step.

> - Exposing basic NetworkEditor operations (connect, create, delete, select, 
> copy, play...), interactive console + completition on ops, processings and 
> ports
> - Full (or better partially) exporting of the C++ CLAM API to allow full 
> programming
> - Being able to implement a processing (there are some real-time constraints 
> not solveable with python)
> - Control scoring, that is (a processing) sending controls to processings of a 
> network according to a python script similarly to csound scores.
> - Descriptors manipulation, for the annotator, that is accessing descriptors 
> for example, as numpy arrays and performing operations on them to get other 
> descriptors.

For me two options would be interesting:
1. exporting of the C++ CLAM API to allow full programming
2. use clam processings and processing-data with python/numpy

In case 2 it would be nice if one gets the processing data in python as 
numpy arrays, so it could integrate in the whole scipy/numpy world.

What I am not sure ATM is how this should work for your dynamic 
attributes - did you already think about that ?

 > My advice is that you should contrast your ideas with ours in a IRC chat.

Yeah ATM I'm not so much on IRC because of my master thesis - but I will 
go online in the next days ;)


More information about the clam-users mailing list