[sdiy] Fixed Filter Bank demo

[Ian Fritz]

OK as far as it goes. But what would you use for “THE impulse response” of, say, a bassoon? Seems to me there is a different one for each fingering. At least that’s what I remember of how the simple physical modeling is done.

I find that using the published formant frequencies for woodwinds helps with getting more natural sounds, but that is only one component of the whole story. Most important is producing natural-sounding transients.

Ian


> On May 23, 2022, at 9:23 PM, brianw <brianw at audiobanshee.com> wrote:
> 
> One technique for modeling acoustic instruments like violin is to obtain the impulse response for the body of the instrument, without the strings attached, and then synthesize the strings separately before running the sound through a convolution process that uses the impulse response to impart the sound of that specific instrument. Folks have painstakingly measured the impulse response of a Stradivarius body in an attempt to capture that sound and reproduce it. There are already fairly lifelike string synthesis algorithms, so imparting the sound of the wood makes the result even more realistic.
> 
> My understanding is that this is what the old preset synths were doing, but with hand-tuned filter designs rather than math and processing.
> 
> Convolution can reproduce any linear filter, and you can even build arbitrary filters by designing the filter response in the frequency domain before using the inverse Fourier transform to produce an impulse response.
> 
> The reverse is also possible. Starting with the impulse response, you can use the Fourier transform to convert to the frequency domain. This will show the frequency response of any arbitrary system. To recreate this ’sound’ with a filter bank, you’d need to manually pick the most prominent peaks, one for each filter channel that you have, and then you can manually dial in the amplitude (and resonance) settings of your filter bank to match.
> 
> Summing the outputs over all notes would give you an approximation of the impulse response. An impulse response inputs all frequencies simultaneously, all at equal volumes (or a sweep is done to avoid overload and distortion), so it would a lot more consistent than playing each note one at a time (given the challenge of playing each one at precisely the same amplitude).
> 
> Basically, the two techniques are probably very similar (IR versus summing all notes). Given the difficulty of ‘inputting’ the impulse or sweep (speaker? transducer?), it’s perhaps a lot easier to sum all notes. The mapping from frequency response to knob settings is going to be approximate anyway, especially with a fixed filter bank compared to a parametric (like the Polymoog Resonator). I’m not aware of any tools that would directly tell you where to set the knobs, but you can probably figure it out by looking at the frequency response and ‘eyeballing’ it.
> 
> Brian
> 
> 
>> On May 23, 2022, at 6:57 PM, Ian Fritz <ijfritz at comcast.net> wrote:
>> As I understand it, formant responses are usually obtained by meauring the sum of outputs over all notes of the instrument. I’m not sure how you would use impulse responses for this.
>> 
>> Ian
>> 
>> On May 23, 2022, at 7:50 PM, brian wrote:
>>> If there is a collection of Impulse Response files for various acoustic instrument bodies, then a frequency analysis of the body IR would give clues for the strongest resonances. Of course, obtaining the impulse response for an instrument body (independent of the strings or reed or other usual input) is probably quite challenging, and I am not currently aware of any such database.
>>> 
>>> Another source might be schematics for those old preset organs that have selector switches for various - usually orchestral - instruments. If they document the schematic thoroughly, then the frequencies and resonances should be apparent. I assume that these would all be fairly coarse, at least due to tolerances on passive components, but surely they’re accurate enough for those old synths.
>>> 
>>> I wrote an AudioUnit plugin that offers the set of filters and controls of a Polymoog Resonator (but it’s ‘ancient' - only 32-bit), so I am also interested in sets of up to three resonant frequencies that would suggest real instruments because these could be added to the plugin as factory presets.
>>> 
>>> Brian Willoughby
>>> 
>>> On May 23, 2022, at 4:34 PM, Ian Fritz wrote:
>>>> Tim —
>>>> 
>>>> I’ve been working on this a bit. I’ve ported my 36 stage hardware filter bank into Reaktor and am using it with my Reaktor ensemble “Winsynth”.
>>>> 
>>>> Surprisingly, there doesn’t seem to be a table anywhere to give a lot of information. I have a textbook that gives coarse data for a few instruments, and a small table taken from a German text.  I seem to remember another old source but cannot chase it down.
>>>> 
>>>> Anyone else?
>>>> 
>>>> Ian
>>>> 
>>>> On May 23, 2022, at 3:05 PM, Tim Parkhurst wrote:
>>>>> Nicely done!
>>>>> 
>>>>> Question: is there some sort of chart or reference guide to mimicking the resonant peaks and valleys of various acoustic instruments? Just going by ear could take forever (or lead down many blind alleys), so some sort of guide for filter banks or EQs would be extremely helpful. A similar guide for the Polymoog Resonator would also be a valuable resource. (Note: I think the Cherry Audio Polymode lets you route other audio in your computer through the Resonator, so that could be fun for those who don’t have the actual hardware)
>>>>> 
>>>>> Tim (resonant peak geek) Servo 
>>>>> 
>>>>>> On Mon, May 23, 2022 at 9:36 AM Ingo Debus wrote:
>>>>>> Hi,
>>>>>> 
>>>>>> here’s another short demo, this time of the Fixed Filter Bank I built 12 years ago.
>>>>>> https://youtu.be/feo7DB85SBA
>>>>>> 
>>>>>> Anyone remember my Fischertechnik coil winding machine?🥸
>>>>>> 
>>>>>> Ingo
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20220523/31203618/attachment.htm>