Well, I read through the entire troubleshooting FAQ, but I don't see this, so, here goes:

When I have the Spectrum Analyzer up on screen, only about 3/4 of the display moves with the music. The display is left justified... Now, this may just be showing me that anything over whatever cutoff point isn't in the music. I figure it might just be the high end loss on the MP3's... However, I used the Aux in, and put some CD's through it to the same avail. Any input on this? I mean, it's certainly not hindering me... I like hula hoops visualization more, but the analyzer is good to make sure the music isn't clipping. Thanks for any help you can give me.

This is simply the nature of the human hearing spectrum. The lowest frequencies are the ones best heard, and the ones most saturated by modern music.

You might think that a high note hit by a singer should show up on the high end of the scale, but actually even when a singer hits a high note, it's generally towards the left or middle of that scale. You'd be surprised how low on the scale the human voice registers.

Only the highest frequencies of cymbals, certain kinds of rhythm instruments like castanets, and certain filter sweeps on synthesizers will make the upper registers move visibly on a spectrum graph. The upper spectrum is active at all times, it's just not active at a very high volume.

Another factor is your MP3 encoder. Most encoders will roll off high frequencies above 16-18k because it makes the file easier to encode. Most folks don't notice the rolloff unless it's too overt. So a lot of the high frequencies are lost on MP3 encoders. But even if you use an encoder that encodes the highest frequencies (or you play WAV files), you will still not notice anything on the right end of the scale very often.

I have been meaning to ask Toby for a "human hearing scale" spectrum display. One that stretches out the lower frequencies, and squishes the higher frequencies over to the far right end of the scale. That way, we'd see more detail in the range that tends to be the most active.

Thanks, TF.

I thought this might be the case (as stated in my original post) I just wanted to make sure that I was correct. I'm in agreement, a "filled out" spectrum analyzer would be sweet, not necessarily as true, but for what I'd be using for, definately more useful.

How squished are these displays already? I tend to find for most of my music I get a fairly full screen - nice and blue at night. For heavy rock, metal and industrial stuff the frequency coverage is just about right. The log displays are definitely better than the plain scale, though.

I'm only referring to the strict linear spectrum analyzers in this discussion, not the event view or log displays. In other words, Spectrometer 16/32/64/128, StereoSpec, etc. As far as I know, their scales are perfectly linear, and for most music, you'll see that they are "left heavy".

For all other visuals, they seem to be properly compensated so that they look good.

Now that I think about it...

Toby, are the "Log" displays already the other visual I was asking for? I thought that "log" meant that the "bounce" response of the bands was logarithmic. But now that I look at them, is the frequency coverage logarithmically compensated, too?

The Aux input is no better a test than an MP3 as it is sampled at 29,400 Hz which gives you a top frequency of 14.7Khz, even lower than the MP3 cutoff.

The logarithmic spectrum analysers (SpectroLog 128 and family) are all corrected with the base of human hearing curve. This effectively reduces the low and high frequency contribution and boosts the mid-range frequencies. The curve is widely available for download and analysis but I can't remember where I found it.

The region in which the curve boosts the signal is in fact the predominantly in the human voice frequency region.

The logarithmic ffts are logarithmic in two dimensions whereas the original Spectrometer visuals are logarithmic in only one dimesion.

Decibels is a logarithmic scale of loudness and both types of fft display their results in decibels, therefore both use a logarithmic vertical scale.

The Spectrometer visuals employ a pretty much linear frequency scale horizontally. This is fairly easy.

The logarithmic fft visuals use a cunning algorithm developed by myself, John Graley, and John Ripley. It uses an a recursive undersampling technique to produce 10 or so input samples and generates some 15 individual transforms from this. The results are then combined to produce the final logarithmic fft. We are extremely excited by the algorithm as we have never seen a high resolution logarithmic fft anywhere, not even in Winamp. With conventional linear transform techniques one would need to generate a 16,000 point fft in order to get the same resolution (this would consume vast amounts of CPU). The trick we have developed gives us the same results at a fraction of the cost. The Rio Central employs these ffts and displays up to 480 point log ffts.

Toby

This is totally cool, Toby, thanks. I didn't realize that the "Log" visuals did all of that.

So... DeadCow, there is your answer. If you want to see a spectrometer that matches your perceived hearing response to the music, select one of the visuals with the word "log" in the title.

Hey, Toby, you know what would be a fun variation on the Log visuals?

You could have a small text number flash and then fade out for some of the highest peaks. For instance, you might see a big peak appear on the spectrum, and at the top tip of that peak, a little number would appear on the screen showing the exact frequency number of that peak. Then the number would fade out or dissolve while the music kept playing. Then when another really big peak appeared, then another number would flash at the tip of that peak, and so on.

You could play around with the sensitivity of the peak detection and the factors controlling the supression of new numbers until you got a system where the numbers appeared in a pleasing fashion instead of crowding the screen with numbers.

Maybe I'm just a strange tech-head who'd like to see the exact frequencies of certain peaks, and this is a silly idea... If so, please pay no attention to me...