Question about Spectrum Lab

That's likely a setting that is off somewhere is my guess. Some kind of digital noise created in the audacity file (I'd guess it's something between the sc and audacity perhaps a sampling rate mismatch)?

A couple of things.

The abrupt start and stop of your signal creates frequencies other than 50Hz. The strongest is at 5Hz because the total signal lasts .2 sec. Those extra tones modulate the 50Hz tone and create the sidebands above and below 50. Note the approx 5Hz spacing of the sidebands.

I don't have any experience with Spectrum Lab so I don't know why it's creating those odd order distortion products centered at 150, 250, etc. Maybe it's clipping? Try generating a signal with lower amplitude and see if those go away.

Here's what it looks like in ARTA. I added a 1 sec silence after the .2 sec signal to give the FFT enough samples to work with. I imported the wav as an impulse response, gated it to .2 seconds, and then generated the frequency response. Other than the log frequency scale, I'd think you should be seeing something similar in Spectrum Lab.

Right. You get sideband every 2.5 Hz (1/.4). Those sidebands are real, strictly a result of the shape of the signal you are analyzing. I don't know how you got those curves with no ripples but they aren't real. But then I don't know jack about SL.

About a kick drum, it's a broadband signal. It has a loudest fundamental but it also contains tons of other frequencies. And yes, clipping creates odd harmonics wherever it happens in the signal chain -- mic, amp, sound card, software, whatever.

I doubt it was Spectrum Lab clipping. It was more than likely the audio encoder in Audacity adding the harmonics.

Check out the raw file Audacity output in Cool Edit to check the wave form out.

I had the same problem when doing my test DVD with sine waves above ~-6dB. The input wave was fine, but the wave to Ac3 (not using Audacity) encoding added harmonics.

The best free/shareware to do this type of thing is the old NCH Tone Generator and/or Cool Edit 95 or Pro. Audacity will only give you a headache tracking random issues.


Edit: adding a little fade in/out at a 1/4-1/2 wave will help some of the issues here.

 

NCH Tone Gen.zip 89.4482421875k . file

I just bought NCH tone generator a few months ago. It can do some cool versatile things and I've not had any problems with it.

FWIW, I generated my wav files in Audacity with no clipping even at gain = 1. What happens if you send them through a sound card, I don't know. That would probably be a setup issue. I don't know why Soho had problems with a straight software conversion (no sound card.)

100ms


100ms zoomed in



3seconds

Ricci,

That is the best color map I have seen posted.

If you decreased "C" a little (to make 0 white, instead of going pink-white-pinkish/purple) I would say was perfect.

I never meant to imply that Audacity would clip the signal. Only that I would look there (first and then elsewhere) before determining that SL was clipping things, or at fault from just this.

I haven't tried Audacity in years, as it failed my accuracy testing, and I stuck with Cubase, Sound Forge, and Cool Edit. At that time it added some aliasing, and a lot of the plug-ins added harmonics IIRC. It was unable to retire CE. This is why there is a negative Audacity slant in the former post, in this test tone thread.

If you play around with different audio formats and encoder software you will find that a lot of them add extra info to a wave if you go back and compare. With sine waves a lot of the time you can hear it when played back to back. With music however, it isn't a big deal. You can almost never tell. Nothing out of the ordinary here.

Audacity is great for it's intended purpose, but I would be worried about it before Spectrum Lab.


What I don't understand is why you are using such high levels in some of those graphs. The 50% pic is ~+8dB, and the 100% is off the chart. Did you calibrate the line in and outs? I would really check there.

Since Ricci added the SL, here is the FR of a 5 cycle, 50Hz wave at 0dB.

 

ok.zip 6.896484375k . file

I see what you are getting at now. I've been wondering the same things. Is the side band content real? It certainly looks like it is.

For some reason I keep going back to Tom D. talking about transients having content that covers a huge spectrum that rolls off on each side of the fundamental.

OK, lets see...

What you are seeing with sine waves is more a product of the way the graph is produced and the wave is simulated than anything else. As catapult stated earlier. It is the way the the start and cutoff of the wave is calculated. This is why when the start and cutoff is far enough apart you get the straight line between the blooms. If you play different different areas of the wave you can get different results.

This is why I included a pic at two different FFT windows. The peak area is all that matters here, all the rest might as well be noise.

With real waves it gets complicated. Musical instruments play a fundamental frequency, but the harmonics are what make everything sound different. This is what a majority of timbre is. In some cases the harmonics are louder than the fundamental. This is called natural frequency(ies), over tones, or formant. This is a different phenomenon from what you are seeing in the sine wave graphs.

A kick drum (any drum) will not have a single fundamental. Based on where it is struck you will get several modes across the surface of the drum, a closed end air column resonance, attack, and a shell resonance.

Look at his "close mic kick bass detail" pic. There are peaks at 40, 55, 70, 80, and 115Hz the way Ricci has his drum tuned. Now if you go hit some sound mixing sites you will see that the classic rock sound is an added EQ emphasis at 80-90Hz, and the new harder sound is EQed to emphasize 40-50Hz. The area below would be rolled off. The key is where you want the bass and drum to be placed. The traditional way is to have the bass be the low end, and the drum is for impact. The new age has the drum defining the low end, while the bass is used in a more melodic way, as opposed to just adding balls. If he expanded the graphs you would see peaks all the way up into the thousands.

There is zero time spent on accurately representing an instrument in mixing. The only thing that matters is how it sounds.


Yes and no. To do it exactly yes, but as with everything there is a point of diminishing returns.


I would say it is more of a physical buildup with the surface nodes becoming randomized to a point. Think of a sink of water. Take your finger and tap the surface crisply one time and watch the wave pattern. Now do it over and over again moderately, and watch new patterns slowly take form. Change tempo and they will fall apart, and then form up again. This is what is happening to the drum head, and I think would cause the effect you are talking about.

if you take the same wave I attached to post 15, and fade in and out during the first and last cycle you get this.

I used a -45dB starting and ending point.


I uploaded the wrong pic. This is the corrected one. I didn't do it at 0dB so it is ~6dB lower than the previous one FWIW.

Soho,

Don't you use NCH? I remember that it could add a fade in and out to the signal but I could not seem to find that option last night. I was going to try it and see what the difference it made was.

OK, last random post here.

I think part of the problem is the range most people use on these graphs. Does that nice bloom -40dB down from the peak mean anything in the real world?

Here is a test.

First off SL should be calibrated just like any other system. Using a know signal you should adjust the line levels until the SL graphs pretty closely to it. Since we are using generated files here make one at -10dB, and calibrate to that.

Create a wave containing three sine waves, as this is mostly an LFE thing let's go with 15, 30, and 60Hz. The first wave should have all tones at -5dB, be a few seconds long, and be faded in/out from -70dB. In each consecutive wave should have 30Hz be -10dB from the last.

The test will be to tell at what point 30Hz becomes irrelevant to your ear. If would be more relevant if you made several runs with different tri-tones, but one should be a start. 30-60-120Hz would be nice, as well as randomizing the variable tone.

I used the old free version, and it doesn't have fading capability. I used Cool Edit for the fading, and to make sure it was 5 cycles.

I can post the wave for you if you want. Here it is. I didn't bump it up to 0dB this time. I also used linear fading, log would look a little different.

 

Ok3.zip 4.9814453125k . file

OK, I got done watching Avatar and I was bored.

Here is a 15-30-60Hz masking test. I went from 30Hz at 0dB to -40dB in 10dB hops.

All files are within 1dB of each other peak wise, with the highest being -.3dB. I leveled with RMS which is at -4.25dB for all.

Honestly, I can't tell a difference between -20 and -30. This means the cutoff here is probably a little north of -20dB. EDIT: with headphones, it is to late to rattle the windows with the wife in bed above me.

I might try it with the tones placed further apart.

http://rapidshare.com/files/381014342/masking_test.zip

15-30-60Hz just goes good with this image from wikipedia.
http://en.wikipedia.org/wiki/File:Audio_Mask_Graph.png

Really? Hmm.. 0 to 10 should be no problem even at low volumes. 10 to 20 as well at moderate levels.


This shows there is a problem. You levels are off. I have the fix.

First off, set up SL. Then run the 0dB wave in a loop in another program. Make sure all EQ settings are zeroed, or turned off.

Set the wave player, master, and wave (in advanced volume controls) sliders to ~75-85%. This is probably what you will get.


Now go to turn your line/mic-in slider all the way down. While watching the spectrum graph, raise the slider until the other peaks start showing up. Then back off a little. Like this:


If you go to far you get this again:


Now all three peaks should be pretty even, just not at 0dB. Go into the SL Spectrum options and adjust the Offset until the peaks are at 0dB like they should be, and you are done.



Note: this was done with a Gen 1 Audigy SC, so the S/N isn't great.

This is what I get for
0dB


-20dB


-40dB