[Clam-devel] Spectral Analysis/Synthesis problem

Xavier Amatriain xavier at amatriain.net
Fri Nov 30 03:15:29 PST 2007


I am surprised that with those settings all you get is a "small glitch". 
Is that with the filter
active? I would predict that as soon as you start doing noticeable 
transformations to the spectrum
with those settings you would have more problems including a sort of 
phasiness. But if not, let me
know. My advice:

- Use an analysis hop size of at most windowSize/4... that is 64 in your 
case
- Use the same analysis and synthesis hop size (again 64)
- In general an FFTSize of 226 is really, really small to do anything 
meaningful but
it might work decently on a 3Band filter so you might want to keep it 
this way to
reduce latency.
- As Pau suggested, the "None" window is not the most advisable (maybe 
the least) because of
the characteristics of its transform. Again in a simple filter and if 
you don't go to extreme settings
you will get by with it but I don't see any reason you should want to 
use it instead of a more reasonable
window such as BH92,

X

Pau Arumí wrote:
> Probably Xavi will have more insight about these issues.
> But the first thing that comes to my head is: it seems to me that you're
> applying a rectangular window (type=None). Have you experimented with
> other window types?
>
> Pau
>
>
> El dv 30 de 11 del 2007 a les 15:29 +0800, en/na globot va escriure:
>   
>> hi,
>>
>> I am sorry to disturb you "again" but here is my problem :
>>
>> I use this kind of network : 
>> SpectralAnalysis--->3BandFilter--->SpectralSynthesis
>> it work almost perfectly (with jackd buffer set to 64) but there is one 
>> small glitch.
>>
>> it seem that every "SpectralAnalysis.HopeSize" there is a peak appearing 
>> in the synthetized sound.
>>
>> here i put a example of what i got with those settings :
>> - analysis.HopeSize = 128
>> - analysis.ZeroPadding = 0
>> - analysis.SamplingRate = 48000
>> - analysis.FFTSize = 256
>> - analysis.WindowSize = 257 (type = None)
>>
>> + synthesis.ZeroPadding = 0
>> + synthesis.SamplingRate = 48000
>> + synthesis.AnalWindowSize = 257 (type = None)
>> + synthesis.HopSize = 64
>> + synthesis.SynthWindowGenerator.size = 129
>>
>> PS : If i modfy the params to have FFTSize = 1024 so HopSize = 512, the 
>> peaks will appear evry 512 sample.
>>
>> Any idea what to do ? this is my last big problem... And i have 3 days 
>> to solve it :(
>> is it possible there is a small problem in SpectralAnalysis/Synthesis 
>> class that would cause this while calculating first or last samples ? or 
>> can this be solve buy changing parameters ? (i didn't find any suitable 
>> combination)
>>
>>
>>
>> _______________________________________________
>> Clam-devel mailing list
>> Clam-devel at llistes.projectes.lafarga.org
>> https://llistes.projectes.lafarga.org/cgi-bin/mailman/listinfo/clam-devel
>>     
>
>
> _______________________________________________
> Clam-devel mailing list
> Clam-devel at llistes.projectes.lafarga.org
> https://llistes.projectes.lafarga.org/cgi-bin/mailman/listinfo/clam-devel
>   






More information about the clam-devel mailing list