Over the past few years I've tuned a handful of speaker systems using mostly shelving type filters. In several cases I've done this directly with DSP Studio in JRiver MC. You might be wondering why would I use shelving filters. Or you might be wondering what a shelving filter even is.
Shelving filters are very simple. Think of the frequency response of a system as a straight line from left to right. The lows are on the left, the highs are on the right. A shelving filter pushes one end of the line up higher than the other. So a high shelf filter at 5000 Hz, would push up all the frequencies from 5000 Hz up to 20,000 Hz up higher than the rest. Or pull them down if you specify a negative value.
A high shelf filter that's -6 dB at 3000Hz pulls all frequencies above 3000 DOWN by 6 dB. These are called "shelving filters" because the frequency response becomes a step or a "shelf". It shifts everything up or down all the way to the end of the line. A picture would probably help. Here's one I found:
http://www.hometheatershack.com/forums/attachments/electronic-processing-equalization-devices/8927d1219189200-sub-eq-room-calibration-equipment-question-shelving-filters.jpgSo that's what a shelving filter is. But why would I want to use these instead of parametric EQ bands? The simple answer is that shelving filters are good for sonic sculpting. They are good for pushing around the frequency response of your system in medium to large "chunks". Parametric EQ bands, even when they are made wide with a low Q are still rather narrow sonically.
What I find with many systems is that they have obvious problems that need to be corrected by moving a lot of the frequency spectrum around at once. Sort of like traditional bass and treble controls. But much more precise and much more powerful. In fact, the last time I did a really successful tuning on a high end car audio system, all I had to work with were parametric EQ bands. But the car had such extreme frequency response problems, that we ended up overlapping parametric bands so we could affect large smooth chunks of the frequency response all at once. The result, graphed on the computer interface looked quite a bit like several shelving filters. I don't think we made any narrow band changes. The result was *dramatic* in a good way. If we had access to shelving filters in the device we were using to tune, it would have been faster and easier to get to the same point.
As an illustration of how I use these let's look at my desktop speaker system and how I've tuned it. My current desktop speakers are an older set of satellite speakers with their own matched passive subwoofer. The 10" bandpass subwoofer has passive crossovers built in to it and it sends the mids and highs to the satellites, which have 4" mids and 1" cloth dome tweeters.
This setup was designed to be a home theater system, or just a somewhat compact set of full range speakers to be used in a normal sized room at a normal listening distance of 6 to 15 feet. This last part is the problem: The distance. Have you ever gotten very close to a set of speakers and noticed they they sounded much brighter or more detailed? That's because highs fall off faster with distance than mids or lows. So, a speaker that sounds really balanced at 8 feet, is going to sound very bright (lots of highs) at 2 feet. Guess how far my speakers on my desk are from my ears. You guessed it! About 2 to 2.5 feet.
So, these really decent speakers are nearly unlistenably bright at desktop distances. Shelving filters to the rescue! My first goal was to reduce the high frequency response. So I made a negative high shelf filter to move down the highs. I started at around 3500 Hz and listened. Then I moved the frequency around and kept listening. What I figured out was that the high boost on these speakers starts rather low. I settled on 800 Hz as my knee frequency and brought everything from 800 Hz and higher *down* by 3dB. Which really helped. The bottom ends of voices started to have a bit more body. But there were still too many highs. I moved up a few octaves and settled on 3500 Hz and made another high shelf: -4 dB .
Astute readers will have realized that this means that everything from 3500 and higher is actually down by -7 dB because the two shelves add together. The frequency response graph, which was originally a flat line now has two steps down in it along it's path to the right.
Now the sibilant parts of voices was vastly reduced and starting to sound very natural. But there was still too much sizzle on the top end of voices making them sound artificial. Just a little adjustment: High shelf -1 dB at 5000 Hz.
By this point things were sounding really nice. But I noticed that cymbals and other very high frequency content had lost a lot of it's "spit and sizzle". They sounded overly muted and lacking that high end sparkle. Again I experimented, but settled on another high shelf, but this time a BOOST: High shelf, +4 at 10,000 Hz. This brought the sparkle back without making voices sound weird, or tearing my head off with the treble. Just some pleasant highs.
There was still something wrong in the vocals, but it was subtle. I played around with a single parametric EQ band and moved it around a bit until I tamed it: frequency of 1700 Hz, Q of 1, -2 dB .
Note that I played around with both the step frequency and the amount of cut each time I made these changes. It took experimentation to figure out each one. I spent around 2 hours making these adjustments initially, and then over the next day made some minor tweaks until I was satisfied.
Reading through this, it might not be obvious why I'm turning everything DOWN and not turning anything UP. In the digital domain, it's normally best to cut, rather than boost, because if you boost, at some point you can over flow the digital signal's dynamic range and cause clipping. So I prefer to cut, cut, cut, and then, if necessary, add back just a bit to bring the overall signal level up closer to where it originally was. In my case I did just this and added +3 dB of Volume. I probably could have added more, but I wanted to be conservative to account for many different types of music (and corresponding frequency blends).
The effect of these adjustments was again, DRAMATIC. The bass is much more prominent and punchy. But not "one note" or "boomy". Because I didn't boost ANY bass frequencies. I just cut everything above 800 Hz, which brought up the bass, the mid bass, and the bottom end of the vocal range. Voices on these speakers originally sounded thin, bright, and "biting". Vocals now are rather natural and balanced in my opinion. The highs are just about right, though some would hear these and think they they lack high frequency detail. I did that on purpose in order to make these speakers non-fatiguing. I find extended "detailed" highs to wear my ears out very quickly. With these speakers, tuned this way, I can listen for several hours, at a moderate to low volume, and be very happy doing so. Which is the whole point right?
I'm guessing that some reading this are thinking that doing what I've done has completely destroyed the sound of these speakers and that my efforts are misguided at best. You're certainly entitled to that opinion. I used to share a similar opinion about equalizers in general. My acid test for equalizer users is very simple: Do you ever touch the eq between songs? If your answer is "of course, I change it for almost every song!", then your EQ is misadjusted. EQ *can* be done right. When it's done right, you won't want to change it for every song. That's how this system behaves for me.
Finally, I think that this process is fairly similar to how speaker designers design and tune speakers. Moving entire sections of the frequency response up and down is done all the time in speaker design. Either by selecting drivers with different efficiency ratings, or by adding padding to the passive crossovers. It's fairly typical to have pads in the tweeter circuit to bring it down in level, as the tweeter tends to be the most efficient driver in a multi-way system. I'm doing somewhat similar adjustments. I just happen to be doing it in the digital domain with more precise controls.
This is my new tuning method of choice for speakers and headphones. I think it works quite well.
Thanks for reading.
Brian.
Author
Topic: Sonic Sculpting Speakers with Shelving Filters (Read 13372 times)