Forums > MaxMSP

Vocoder Freq. bands without FFT

January 8, 2007 | 7:42 pm

Is there anyway to divide a signal up into seperate frequency bands without using a fourier transform? I’ve been trying to do this using biquad~ but the rolloffs aren’t steep enough so each band is bleeding into the ones next to it. Are there any filters that have extremely steep freq rolloffs, and no resonance?


January 9, 2007 | 1:35 pm

Try using several biquads in a serial chain. If i’m not mistaking biquad has a 6dB rolloff. by using 4 in a row you have a 24dB filter. You can use spectroscope~ to see the bandwidth and steapness of the filter.

You can checkout the vocoder patch by Marcel Wierckx in the MaxMSP examples.

Goodluck with it.


January 9, 2007 | 2:23 pm

On 08 Jan 2007, at 20:42, Chris wrote:

>
> Is there anyway to divide a signal up into seperate frequency bands
> without using a fourier transform? I’ve been trying to do this
> using biquad~ but the rolloffs aren’t steep enough so each band is
> bleeding into the ones next to it. Are there any filters that have
> extremely steep freq rolloffs, and no resonance?

there are ways to do this, but it’s probably not very efficient in
the end. that’s why ffts are so prominent…
you’d need to look into FIR filters if you want no resonance.
to obtain steep rolloffs, you’d need *a long* filter kernel.
to experiment you could have a look into linear convolution using the
buffir~ object.

there was a similar thread on the list not long ago.
search for "creating ideal filters".
i’ve built a patch that outlines one possible way of creating custom
FIR filters.
it’s still online: http://www.esbasel.ch/Downloads/FIR_Filter_Design.zip

maybe this gives you some ideas on how to go about it, or at least
where the difficulties are.
i guess it’s overkill for a classic vocoder implementation.

volker.


Viewing 3 posts - 1 through 3 (of 3 total)