Integration of Rasta, MFCCs

19 views
Skip to first unread message

David van Leeuwen

unread,
Oct 7, 2014, 3:45:50 AM10/7/14
to juli...@googlegroups.com
Hi, 

In trying to submit my own versions of SignalProcessing and Rasta I was pointed out to that fact that DSP was already in METADATA.  

So if everything is all right with you, I'd like to compose a few pull requests to incorporate some stuff I've done for speech feature extraction (we discussed this about a year ago) into DSP.  

I managed to get almost exactly the same result with DSP's spectrogram() as with my version specgram() which was modeled after Octave's implementation.  A small difference is that with my code, the energy for 0 and FS/2 were lumped in the same bin whereas in DSP these are separate, which probably is more correct.  So that means that the wrappers I use could work seamlessly with DSP. 

I understand from the Julia METADATA maintainers that the tendency is to have fewer larger packages rather than more smaller.  This would mean that I might want to integrate both Rasta (auditory spectral analysis) and MFCC (standard feature extraction) into DSP---do you feel that these still are fitting the realm of DSP?

Cheers, 

---david

João Felipe Santos

unread,
Oct 7, 2014, 8:50:05 AM10/7/14
to David van Leeuwen, juli...@googlegroups.com
Hi David,

it's great to know you are going to push your packages to METADATA! I think you did a good job on them and they definitely are going to be useful to other people.

What has been discussed previously by the Julia DSP community is that DSP.jl should have only "general-purpose" DSP functions. The only things you have in SignalProcessing that we don't are levinson and toeplitz, so we could add those if they are not derived from the Octave implementations (otherwise we would have to change our license).

Something like RASTA and MFCC extraction should definitely live in a separate package (or packages, maybe?). I am also working on some separate packages related to speech/audio research (auditory filterbanks, speech quality metrics) and will keep them as separate packages as well.

Cheers

--
João Felipe Santos

--
You received this message because you are subscribed to the Google Groups "julia-dsp" group.
To unsubscribe from this group and stop receiving emails from it, send an email to julia-dsp+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply all
Reply to author
Forward
0 new messages