[Clam-devel] Re: Attacking your core goal

[Jun Jun]

The wiki is on again. :-)
The "Detrended Fluctuation Analysis" is a strong feature about danceability. It was refered in the mail before the last one in this thread.
It originates from fractal ananlysis and reveals correlations within data series across different time scales. In the implementation, there are some parameters (e.g., the size of MIN time scales, the size of MAX time scales, the order of the regression fit) which should be configurable. I'd like to know whether the subclasses of ProcessingDataConfig or ProcessingConfig are those I should look into.

> Normally we do unit tests (white box) for funtional components we can now the 
> result, and back2back for complex computations to assert that the results are 
> not modified without noticing. In your case, and for chord extraction, we 
> should also add some 'fitting' tests (comparing with a ground truth) to 
> support b2b when the results differ and it is suposed to be an enhancement.
> 
> Is that something i could help you in any way?

It helps, thanks, and more demonstrations of documents, code,etc., could be further helpful, thanks again.

>> Given these problems solved, the next plan is to extract more semantic
>> music descriptors (strong related to danceability) based on some exsiting
>> CLAM descriptors including Onsets and BPM.

> It sounds funny ;-) Just keep us informed and we'll try to help you as much as 
> we can. Are you planning to do that development on clam svn?

Yes, I am, but please tell me if such development is not appropriate for clam. (due to the complexity? or the extractor should be separate?) In my opinion, yes it's natural that users annotate a file by hand with a high semantic descriptor, but a content-based extraction of the descriptor is also reasonable, even if just for referrence.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.clam-project.org/pipermail/clam-devel-clam-project.org/attachments/20081216/87c48195/attachment-0004.htm>