Spectral Analysis

[nyjm]

A propos of Manhattan Glutton's comment:

How do I perform a spectral analysis of a song? I looked it up here and I understand the concept, but from the article it looks like there isn't much in the way of freeware out there. Is this a function of some recording/mixing software, perhaps? Like Audacity?

[Billy's Little Trip]

Here are some free and demo anal-izerz. >> http://www.hitsquad.com/smm/win95/SPECT ... /freeware/

The only way I can truly understand something is by using it. I'm going to try one of these now that you've peaked (no pun intended) my interest.

I've D/Ld 3 of them. One for all audio, one for wav and one VST. I REALLY like the discription of this one >> http://www.hitsquad.com/smm/programs/InspectorWIN/
"But wait! Inspector’s usefulness doesn’t end there. Use Inspector's unique alarm feature to specify what conditions are considered adverse, so Inspector can warn you about them."

edit: Damn, the Inspector analyzer free D/L is gone. Drats.

[HeuristicsInc]

Spectral Analysis can also be called a Fourier transform. It involves some signal processing "magic" (I took a whole class on this in grad school and it's pretty arcane stuff). It basically breaks down your time vs. amplitude plot (what you're used to seeing) into its component frequencies (sine waves) plotted as frequency vs. amplitude at that frequency. So your lower frequencies are at the left, generally, and go higher as you move right. Anyway, it may be called Fourier or something else in your software.
-bill

[PlainSongs]

edit: Damn, the Inspector analyzer free D/L is gone. Drats.

Still available here:

http://www.kaosaudio.com/roger-nichols- ... ctor-free/

Looks like a nifty plugin! and manual too.

Spectral analysis is the bee's knees, not just for audio but also for images or anything that has frequencies in it really. You could do a spectral analysis of how often you eat spaghetti.

Also, nyjm, check out ReaFIR as suggested by Adam! here.

[Billy's Little Trip]

Thanks, PS. I'll give it a try and post my results here in this thread.

[Ross]

now that you've peaked (no pun intended) my interest.

must have been intended, since you meant it "piqued" your interest.

[Billy's Little Trip]

now that you've peaked (no pun intended) my interest.

must have been intended, since you meant it "piqued" your interest.

Thank you, but the word play was intended to be graphic meter humor. I forgot, scientists are literalistic.

[jast]

Spectral Analysis can also be called a Fourier transform.

Nerd mode activated
Actually, a standard Fourier transform is rather useless for this. Additionally, there are other spectral analysis techniques apart from Fourier-based ones.
If you go ahead and Fourierify an entire song, you get an analysis over the full time period. The result is worlds apart from the typical frequency analysis things you see; it's sort of a "big picture" analysis. Not too useful for most purposes. Generally you actually want three-dimensional data: how do the amplitudes of various base frequencies develop over time? This is called a spectrogram. The most popular technique to generate one is probably a series of windowed Fourier transforms, i.e. you cut your signal into (usually overlapping) frames and Fourierify each of them, then tape everything together.
I'll spare you the remaining details even though they are incredibly fascinating and everything. No, really...

[jb]

Yeah, but what's the damn POINT of the spectral analysis? How do I use it to make my song better? It's apparently a visual analysis sort of thing, where you make changes to the song to change the peaks and valleys and whatevers of the spectral analysis. But how do you know what's BETTER?

I hate shit that's all vague and "depending on what you want to do" with no concrete specifics or examples. That shit can go to hell. It can go to hell and die!

[Billy's Little Trip]

For the past hour plus, I've been playing with the ones I D/Ld and I'm not seeing how this is going to help me, as yet, if ever. So I'm going grab my pitchfork and torch and stand behind JB with an angry face.

[JonPorobil]

Two concrete examples of things I use it for.

I various songs (including "Back from Juvie,") I've used it to take the wind out of overzealous plosives.

Also, if there's a clip in the mix, spectral analysis can be used to pinpoint exactly which frequencies are overloading, so that you can deal with just the offending frequencies and leave the rest of the sound intact.

I find it's great for spot-EQ work.

[Billy's Little Trip]

I guess I pay a lot of attention to my original recording. But that's not to say I won't find a use for it. Just that so far, it's consuming a lot of my time with no benefit to me. Arrg!

[HeuristicsInc]

The most popular technique to generate one is probably a series of windowed Fourier transforms, i.e. you cut your signal into (usually overlapping) frames and Fourierify each of them, then tape everything together.

yeah, i left out the part about doing it in windows... sue me
-bill

edit: dirty minds could read that above sentence and read other meanings into it.

[Billy's Little Trip]

Voyeuristics Inc.

[JonPorobil]

I guess I pay a lot of attention to my original recording. But that's not to say I won't find a use for it. Just that so far, it's consuming a lot of my time with no benefit to me. Arrg!

Sure, I can see the argument. The first time I looked at a spectral view of a waveform, it confused the bejesus out of me. I also can see the argument that if I'd recorded my vocals right in the first place, I wouldn't need to de-plosive them.

To that end, spectral view can be argued to be beneficial only for demonstrative purposes. For instance, over in the IRC room, Mikelamb was asking about a particular noise in a track, but nobody else in the chatroom could hear what he was talking about. He took a screenshot of the waveform's spectral view, found an orange band in the 10khz range and said "There! That line! What's making that noise?"

Still didn't get an answer, but at least it proved he wasn't hallucinating.

[Billy's Little Trip]

Ok, I see what you mean, Jon. Now that you say that, I had an instance like that and I couldn't figure out where it was coming from. I eventually found it and it was so stupid I could have kicked myself. But that may have helped me locate it, maybe. I screwed with wires, levels, etc etc, for hours before I found it. Also, it doesn't help when you're intoxicated during production. I've made a new rule about that. Recording intoxicated = good. Final production intoxicated = bad.

[jb]

Hmmm, well, I can zoom in on a wav form and see peaks and squash them to order, but I dunno that that's fancy spectral analysis like those plugins do. And yeah, you can maybe spot a problem that you're hearing in a spectral image-- but I wanna know how that helps more than what you're hearing helps.

Practical, concrete techniques trump theory for me where things that I barely understand are concerned. After I learn what knobs to twiddle to get an effect, then I can learn why. I'm not good, and I think most people aren't good, at learning theory and then deductively reasoning out how to apply that theory to the music. At the most, it takes so much longer than somebody saying "put a high-pass filter at about 180 to clear out the mud".

Maybe it's just me.

[Billy's Little Trip]

It's definitely not just you. In my first response to the OP, I linked spectral analyzers and noted that the only way I can understand things like this is by using them. I even played around with them more after my complaint above, but I'm still not seeing how this over use of tech is going to help me more, in comparison to wasting my goddamn time. In the technologies defense, my music isn't very complex, so I'm sure I'm not the target user. Not to mention that the one I was using froze my f**king computer twice. But that's my fault too for owning an older computer, not the programs fault that I don't have the Lamborghini of computers.

[JonPorobil]

Hmmm, well, I can zoom in on a wav form and see peaks and squash them to order, but I dunno that that's fancy spectral analysis like those plugins do. And yeah, you can maybe spot a problem that you're hearing in a spectral image-- but I wanna know how that helps more than what you're hearing helps.

Practical, concrete techniques trump theory for me where things that I barely understand are concerned. After I learn what knobs to twiddle to get an effect, then I can learn why. I'm not good, and I think most people aren't good, at learning theory and then deductively reasoning out how to apply that theory to the music. At the most, it takes so much longer than somebody saying "put a high-pass filter at about 180 to clear out the mud".

Maybe it's just me.

Sounds to me like you're arguing more for what I'm talking about than against it.

Maybe we've got different ideas of what "Spectral Analysis" means?

I use Audition 1.5, which only has linear (as opposed to logarithmic) spectral view. What it does is eliminate the standard volume peaks in favor of color coded patches, essentially creating a "third" dimension, so the x-axis represents time, the y-axis represents frequency, and the intensity of the color represents the volume. So I look at the spectral view, see a dull orange band at 180, and wonder "hey, what's that?" Then I select a rectangular section along that bright orange band, reduce volume by a couple of db, and hear the difference. See this link for more info.

To me, that's not learning some random theory without learning how it applies to the music. That's actually learning how the levels at certain frequency levels effect the music. I've found it very educational, learning this view.

[jast]

I don't use visual spectral analysis much myself, jb. Sometimes I use it to see if anything looks completely wacko, i.e. extreme spikes or something like that. I go by ear the rest of the time.

[irwin]

Jast brings up an excellent point. The spectral analysis tools are great for helping you develop your ear, but the ultimate goal should be to learn to trust your ears. You don't listen to music with your eyes, after all.

So you pull up the spectrogram of your guitar track, and it's really hot around 250Hz, so you cut that, and things sound better. Over time, you start to learn what that sounds like and you can reach for the EQ knob without looking at the spectrogram.

You will never see, "Man this sounds like crap, but the spectrogram looks great!" in the review thread.

Scratch that, you probably will.

[jb]

Now, if somebody went and took one of my songs, ran a spectral analysis and said "you see there, in the whosiwhat, it's too jiggermahootie." Then if they fixed it and did another analysis and said "now look at it. that's how it should look." And if it actually sounded better... then I'd be like "hey man, who's not using spectral analysis? Chumps!"

I'm just sayin'.

JB