reading spectrograms

is it possible to gather enough information from a spectrogram to understand how a sound was made? This is zoomed in on about one second. The bands there are the partials of the sound yeah? And theres a pitch envelope at the beginning of it that goes up an octave then back down. Does the fact that the upper partials are fading before the lower partials implies a filter sweep? I noticed that the partials are not all multiples of the fundamental, which would make them inharmonic I think, do you think these were a part of the original source or added through some sort of effect?

Not sure how easily you could reverse engineer a sound like this. Inharmonic partials could be created any number of ways including sum and difference tones of other partials. The fact that the upper partials fade first is normal though, happens with natural sounds like tapping a wine glass or whatever.

  On 2/11/2012 at 12:34 AM, gritbox said:

 

 

is it possible to gather enough information from a spectrogram to understand how a sound was made? This is zoomed in on about one second. The bands there are the partials of the sound yeah? And theres a pitch envelope at the beginning of it that goes up an octave then back down. Does the fact that the upper partials are fading before the lower partials implies a filter sweep? I noticed that the partials are not all multiples of the fundamental, which would make them inharmonic I think, do you think these were a part of the original source or added through some sort of effect?

Would probably be easier to just listen to the sample and then recreating it

i've done this with speech, and i think it's possible to get a good amount of info about any sound like this. it's pattern recognition so it just takes practice.

i had a phonetic analysis class where we practiced reading speech sounds on a spectrogram. it takes a lot of effort to be able to do this well, but it's not too hard to make rough guesses as to what consonants lead into what vowels based on movement of the first two formants. some sounds are really easy to read; sibilants like "s" and "sh" are probably the easiest because they're some of the only sounds that make a cloud of high frequency noise.

your analysis of that image sounds about right - there's definitely a glide up and down about an octave, and the higher partials fading first is natural like jim said. that lower frequency content that comes in about halfway through looks like a separate generator might be playing along with the first one. what's the source of the sound?

the tuss - rushup i bank 12, that distorted bell sound(s), probably multiple things layered, used in the first half. Ive given up trying to emulate it, it was preventing me from just enjoying the music. The part that gets me is how much texture there is in that quick little attack, fucking good.That particular snapshot is at 1:05 because it was the most isolated i could find it

Six posts and no one posted this ..

Create a program that creates sound based off of a spectrogram analysis.

Is this possible?

  On 2/14/2012 at 3:07 AM, ZiggomaticV17 said:

Create a program that creates sound based off of a spectrogram analysis.

 

Is this possible?

I believe it is. I have read something similar to that before anyway. I'll get back to this convo with citations.

Spear.

edit: actually no- they could easily add that function though...

  On 2/14/2012 at 3:11 AM, StephenG said:

  On 2/14/2012 at 3:07 AM, ZiggomaticV17 said:

Create a program that creates sound based off of a spectrogram analysis.

 

Is this possible?

 

I believe it is. I have read something similar to that before anyway. I'll get back to this convo with citations.

virtual ans has this functionality, but i don't know if it can do very large images. There's also spectrobits, a vst that can let you play images or spectrograms. Photosounder as well

  On 2/11/2012 at 12:34 AM, gritbox said:

i dont know where this is from, buit i can tellby looking at it it was done with some sort of picture to spectral program or maybe even metasynth itself. It's using visual processing effects on the spectrum as evident by the duplicated 'smaller' version of the spectral print going up like that.

are you sure? Cant the 'smaller' versions happen naturally with the right waveform? If you look at a picture of a square or saw you can see the pattern of partials but they are more organized or even an aural exciter that adds harmonics does something similar. I guess both ways are possible, your description sounds like the right method in this case though

http://en.wikipedia....Graphical_sound

http://www.youtube.com/watch?v=Q0vgZv_JWfM

that clip is great psn, i had not seen that.

  On 2/14/2012 at 4:48 AM, gritbox said:

are you sure? Cant the 'smaller' versions happen naturally with the right waveform? If you look at a picture of a square or saw you can see the pattern of partials but they are more organized or even an aural exciter that adds harmonics does something similar. I guess both ways are possible, your description sounds like the right method in this case though

yeah if it's in a recent aphex track, and it's the part im thinking of in that particular track i'd guess it's metasynth or something very much like it. Instead of doing the traditional jungle time stretch stops/cuts he often uses some kind of specteal smearing effect which is very easy to do in metasynth

http://en.wikipedia.org/wiki/ANS_synthesizer

these videos are very interesting thank you psn

  On 2/11/2012 at 12:34 AM, gritbox said:

is it possible to gather enough information from a spectrogram to understand how a sound was made? This is zoomed in on about one second. The bands there are the partials of the sound yeah? And theres a pitch envelope at the beginning of it that goes up an octave then back down. Does the fact that the upper partials are fading before the lower partials implies a filter sweep? I noticed that the partials are not all multiples of the fundamental, which would make them inharmonic I think, do you think these were a part of the original source or added through some sort of effect?

It's somewhat possible, and yes, your statements seem generally correct. It seems likely that this is a harmonic timbre, though, not an enharmonic one... I could be wrong, but that octave slide-and-back at the beginning shows that the second frequency is exactly double the fundamental, the fourth is double that again, and so on... which the numbers on the left seem to agree with from what I can make out. Yes, the upper harmonics fade out first, which as Jim points out is what usually happens in nature. This is almost certainly why low-pass filters are the most common: they sound the most natural, and so pleasing to us. But yes, if it's a synthesised sound, it looks like it's being run through a lowpass filter with a decaying cutoff point... and as it looks like each multiple of the fundamental harmonic is represented (the first ten or twelve, at least), and nothing else, quieter the higher up you go, but not too steeply so, I'd guess it's probably a sawtooth waveform, although I couldn't guess the direction, not that it would be relevant to anything anyway. Although maybe I'm wrong and it is quite a steep falloff, which might make it a different starting waveform, or at least filtered quite low... but it's clearly not a triangle nor square wave because it contains both odd and even harmonics. See those two bursts of absolutely every frequency (over about 2k, at least) over on the left? That'll be something clangy and noisy, like white noise or a snare drum. Which is why white noise is pretty good for the actual snare part of a snare drum. It looks like it's clustering around the 7k mark, so I'd guess it's going through a band-pass filter with a pretty mellow Q. And there's clearly something else going on at the bottom, what looks like a bassline... the way it looks like a regular waveform oscillating, that might insinuate it's FM synthesis, I guess? I've never looked at FM synthesis in a spectrogram before (or any of this other stuff, for that matter), but this makes me want to, it looks pretty cool! :D So yeah, that's my guess... Oh, and the note appears to be roughly A# / Bb 5.

  On 2/14/2012 at 3:07 AM, ZiggomaticV17 said:

Create a program that creates sound based off of a spectrogram analysis.

 

Is this possible?

You can get several. See Windowlicker for an example. You're basically doing the exact same thing, only in reverse, to translate between sound and a visual representation of that sound.

  On 2/14/2012 at 3:01 AM, gritbox said:

the tuss - rushup i bank 12, that distorted bell sound(s), probably multiple things layered, used in the first half. Ive given up trying to emulate it, it was preventing me from just enjoying the music. The part that gets me is how much texture there is in that quick little attack, fucking good.That particular snapshot is at 1:05 because it was the most isolated i could find it

Ack, so those are hi-hats, not snares... ~7k snares, what was I thinking? I forget just how high up such things sound... I thought the main gist of that song was playing an old piano, or samples thereof, and wobbling the pitch of the recording? It had never occured to me that instrument might not be a piano, to be honest...

Anyway, if you want to emulate something, emulate the originality, not the result of that particular burst of originality. :)

  On 2/14/2012 at 9:21 PM, ZoeB said:

 

Ack, so those are hi-hats, not snares... ~7k snares, what was I thinking? I forget just how high up such things sound... I thought the main gist of that song was playing an old piano, or samples thereof, and wobbling the pitch of the recording? It had never occured to me that instrument might not be a piano, to be honest...

there is definite piano running through the track ,aphex just cleverly puts it through a lot of non stock effects, effects you wouldn't find on a (every dj has one) Kaoss pad. Its the same way 'deathfuck' has a very slight wet H3000 eventide pitch delay with a short delay time and high feedback running on the drums, subtle but original and not many people without a vocabulary of every genre of effect what have any idea how to reverse engineer it

  On 2/15/2012 at 9:38 AM, xxx said:

I've noticed that if you take natural sounds like a piano and completely squidge the hell out of them via compression, distortion, etc., you can get some interesting dynamism that you don't get from doing the same thing to a static waveform. I guess that's an obvious statement but the point is that it becomes so divorced from what you started out with that, if you listen to it with "fresh" ears, it sounds so synthetic that your first instinct is to pursue synthetic means to recreate it. Naturally, Rich is probably doing a lot more than going hard on the FL effect suite ( ) but I would ditch the oscilloscope and get to work! find something even more rad in the quest to replicate

Yes, you can get nicely subtly always changing sounds by using acoustic original sources (even if it's just orchestral samples or recordings of household objects) and then processing them like you would anything synthetic. You have all the advantages of a starting waveform that's not just a simple periodic one, plus all the advantages of applying crazy effects to them... somewhat blurring the line between sampler and synthesiser, just as Akai intended, and between synthesiser and effect, which was a pretty fuzzy line to being with anyway. :)

On a slight tangent, I love how the "opening guitar stab" on If You Tolerate This Your Children Will Be Next is actually an electric piano going through an MS20 filter. That's the main advantage of modular and semi-modular gear, it reminds you that anything can and should become a part of your synthesiser... Just experiment with getting disparate and heterogeneous things talking to one another and see what you can come up with!

  On 2/14/2012 at 3:07 AM, ZiggomaticV17 said:

Create a program that creates sound based off of a spectrogram analysis.

 

Is this possible?

Coagula light - http://hem.passagen.se/rasmuse/Coagula.htm

nice, still have to check out this program. Im still waiting for a genius from this forum to make a max/msp patch that lets you do effective sonification using spectral data to jitter animated spectrum and subsequently enabling you to put jitter's video effects on said spectrum spitting the audio back out at the end.

  On 2/11/2012 at 12:34 AM, gritbox said:

is it possible to gather enough information from a spectrogram to understand how a sound was made? This is zoomed in on about one second. The bands there are the partials of the sound yeah? And theres a pitch envelope at the beginning of it that goes up an octave then back down. Does the fact that the upper partials are fading before the lower partials implies a filter sweep? I noticed that the partials are not all multiples of the fundamental, which would make them inharmonic I think, do you think these were a part of the original source or added through some sort of effect?

Yes it's possible to understand how a sound is made this way and recreate it, I do it using Photosounder, it can give interesting results, it's a pretty powerful way to learn how to synthesise sounds/recreate instruments, you can even recreate speech (or even the THX sound lol). This being said I like to recreate them graphically, it's more straightforward (but hey that's just my thing). However I don't see any inharmonic partials really in this sound. In a lot of sounds the upper harmonics fade out faster than the lower harmonics, for example you can see that in a guitar I think.

  On 2/14/2012 at 4:40 AM, Awepittance said:

It's using visual processing effects on the spectrum as evident by the duplicated 'smaller' version of the spectral print going up like that.

Are you talking about the harmonics?? That's just how harmonics look.

  On 2/15/2012 at 11:14 AM, mcbpete said:

  On 2/14/2012 at 3:07 AM, ZiggomaticV17 said:

Create a program that creates sound based off of a spectrogram analysis.

 

Is this possible?

Coagula light - http://hem.passagen....use/Coagula.htm

Coagula doesn't analyse sounds though. That's where Photosounder comes in ;)