Hello,

I'm planning to do a graphic equalizer as fast convolution FIR filter in Pd. Let's say the equalizer has 10 sliders, one per octave. The question is how to translate the octave amplitudes to a smooth frequency curve. I was thinking about something along these lines:

1. upsample an array containing the amplitudes via FFT and larger zeropadded IFFT to get a smooth curve of 1024 points

2. warp the linear curve to logarithmic using [tabwrite4~] interpolation, to get 1024 spectrum points

Not sure if step 1 will work. Before reinventing the wheel, I'd like to know if someone has already built a graphic EQ FIR filter in Pd.

Katja

OK I got the graphic EQ done, but not with zeropadded IFFT. I could have known: that gives a sum of sin(x)/x functions, very smooth but not quite the shape I need. Instead it works with a few extra steps of cubic interpolation. Attached demo patch has a 10 band EQ with -12 till +12 dB amplitude range.

As implemented in the demo patch, the fast convolution filter has long latency. However there is a method to make zero latency FIR filters. I've used it before with multipeak filters:

http://puredata.hurleur.com/sujet-6515-latency-filters-minimum-phase-partitioned-convolution

Graphic EQ with zero latency FIR, that's what I'm heading for. The purpose is to improve the frequency response of unidirectional microphones (with proximity effect) in live performance.

Katja

http://www.pdpatchrepo.info/hurleur/EQ.pd