Can you derive a freq plot from a sound clip?

[JD3971]

Hi all,

Quick introduction - I'm a luthier by hobby and engineer by profession. I read Trevor's books a few months ago and have been doing a lot of thinking. I watched a video of Kenny Smith playing Norman Blake's D-18H a short while ago and though "WOW, I want to build THAT". And that got me thinking... can you use a high quality sound clip of an instrument and use it to figure out the freq plot? I did a quick search on this forum, but couldn't find anything (if this has been discussed before, please let me know!)

I saved the video as an mp3 and ran a FFT on it and came up with a plot, but of course the peaks are on every note Kenny plays in the video as one would expect. And I know that mics, room, mixing will have a part in the result as well, but /somewhere/ in this plot must be the guitar. Has anybody tried to or had any success in doing this?

The song FYR:
https://youtu.be/jyzPvySio-k

The FFT plot for the song:

[johnparchem]

I never determined the resonances from a recording, but knowing the resonances I was able to see them in a plot while playing.

[Jim watts]

I don't have any idea how to do it, but it seems filtering out the frequencies of know musical notes might be a place to star. Or, perhaps set filters that'll only let frequencies pass in the ranges your interested in, minus the musical notes (I'm assuming the instrument doesn't have a natural resonance on a note). Maybe something like 90_110 hz and the 170 - 210 hz. then something for the back. You could probably write an matlab script to do this for you, or use Octave, it's free.
Just thinking out loud here.

[JD3971]

I don't have any idea how to do it, but it seems filtering out the frequencies of know musical notes might be a place to star. Or, perhaps set filters that'll only let frequencies pass in the ranges your interested in, minus the musical notes (I'm assuming the instrument doesn't have a natural resonance on a note). Maybe something like 90_110 hz and the 170 - 210 hz. then something for the back. You could probably write an matlab script to do this for you, or use Octave, it's free.
Just thinking out loud here.

That's similar to what I was thinking Jim - maybe I'll give it a shot. There appears to be a swell on either side of each note peak though which I'm a little confused about (it seems like way too strong of a correlation to be the guitars actual modal freqs).

[North]

Great idea! If you manage to get a method to capture the whole exact spectrum, that would be really, really nice!!

You may at least be able to detect some of the lower resonances if you can find passages with maybe only one treble note ringing, and no basses.

I have done so several times, but it's not always easy to find such ideal excerpts though. And it may be even more difficult if you are interested in one specific guitar only.

Best regards,
North

[johnparchem]

My thought is to collect a lot of samples and then pull out the peaks that show up a lot more than others. There should be resonant peaks off of every impulse. I was watching a spectrum graph in real time as I played. I knew what the tap spectrum looked like and I could see it.

I just thought the mic is right in front of me so ... Here is a sample spectrum playing and tapping. I was playing a solo part on the E and B strings.

Playing:

playing.jpg (56.32 KiB) Viewed 12485 times

Tapping:

tap.jpg (48.15 KiB) Viewed 12485 times

[Jim watts]

John, That's an interesting experiment, Thanks for posting it.
Pretty cool!

[Trevor Gore]

...And that got me thinking... can you use a high quality sound clip of an instrument and use it to figure out the freq plot?...

As John Parchem has said, every played note contains an embedded impulse response. The best way I've found to derive a frequency response curve is to isolate a single note played somewhere high on the fretboard with the longest duration you can find. Using something like Audacity, find that single note in the recording, highlite it and run a spectrum analysis over that sample. You then only have one note and its harmonics to ignore. If you can find a second, similar, but different note you can fill in the gaps. That will usually get you the low order resonances fairly accurately, but it tends to be the higher ones that make the differences you wanted to find.

[JD3971]

Thanks all, that's helpful. Finding a distinct single note tone can definitely be tricky but it's an interesting experiment. Very cool results.