Sealed ESL Design for DIYers

[Altaic]

So, I'm making ESLs (yay) for my room. The acoustics in my room suck, but a discarded roll of wall-to-wall carpet will fix that =)

The main things I worry about these days when designing an ESL panel are bass and arching. So I modified my design (note this is my plan of action; it's not tried or tested):

A = 1" acoustic insulating foam with 1" clearance from frame
D = diaphram
F = frame
G = 3/16" rubber insulator gasket
G2 = 3/16" rubber insulator gasket with two 1/2 inch sections removed opposite eachother
S = stator
C = cloth cover
R = conductor ring with wire attached to the outer edge

-= original =-
|A| |C| |F| |S| |G| |D| |R| |G| |S| |F| |C|

-= new and improved =-
|A| |C| |F| |G| |D| |G2| |S| |G| |D| |R| |G| |S| |G2| |D| |G| |F| |C|

Assembly:
1) Stretch the three diaphrams on three separate bike-tube jigs
2) Coat one of the diaphrams on both sides with Elvamide (http://www.quadesl.org/Area_51/area_51.html or http://www.polymart.com)
2) Spray adhesive on one side of the Gs, G2s, and the conductive ring
3) Attach each of two Gs to a frame, each of the remaining two Gs to a stator (smoother side), and each G2 to a stator (the bumpier side), making sure they all are aligned
4) Attach the conductive to one of the Gs attached to a stator
5) Place the pieces/layers as shown above (without the A and C)
6) Make sure everything is aligned and clamp the frames on each corner (or more; I used 1/8" thick aluminum as my frame, so I'm not worried about the pressure being uneven)
7) Drill a hole in the center of each corner of the assembly, taking care to not to dislodge the clamps
8) Bolt each corner, tightening until the inner side of the stators are 7mm apart
9) Remove the clamps
10) Glue/seal a small hose to each of the openings the gaps in G2 created
12) Fill a balloon full of N2 or your favorite innert gas
11) Kink one of the hoses and slowly pump out the air for one side of the assembly until the diaphrams almost touch the stator (try to get it as close as possible)
12) Attach the balloon full of N2 to the hose you used to pump the air out of the chamber, and slowly/carefully refill the chamber with N2 (you don't want the diaphrams bulging outwards)
13) Unkink the other hose, flush the chamber with N2 and kink both hoses
14) Repeat steps 11 - 13 on the other chamber
15) Remove one hose at a time, filling the hole in the gasket with silicone sealant
16) Let sealant set/dry
17) Cut diaphrams from jigs at the edge of the gasket and blow-dry so the jaggedness goes away
18) Finishing touches go here. For instance the cover cloth and the acoustic foam with 1" spacers, along with a baffle if you so wish, a box of some sort, or a stand/mount

Notes:
- One could attach two valves to each chamber instead of two hoses, and leave them in.
- Flushing the chambers is not necessary, but it will increase the concentration of N2, which will reduce the chance of arching.
- Since the stators are so far apart (most bass ESLs are a max of 6mm) and there are three membranes (3 times the weight), the bias will have to be greater. Hopefully the resultant sound have richer bass while keeping the highs crisp.

Edit: I have a full parts list with online distributors, part numbers, prices, and so forth I can post if anyone wants it. It's costing me around $100 per 18"x36" ESL. It is the original list, but plenty of diaphram and gasket material is already included, so it should work fine.

Enjoy!

 

[MRehorst]

Balancing the gas pressure on the two sides of the diaphragm might be a little tricky. A pin hole in the diaphragm would allow the gas pressure to equalize itself. What will happen on days when the barometric pressure increases or decreases?

I suggest drilling before you attach any diaphragms. The drill is going to drop sharp edged crumbs into the speaker, which can't be good, and operating a relatively large, heavy, sharp, motorized tool within a couple inches of a tightly stretched, 6 um film is a recipe for disaster.

What is the purpose of the acoustic insulating foam on the back of the speaker? I suggest you try the speaker without it first. You'll probably like the sound better...

Is all this really necessary? What problem are you trying to solve with all this extra effort/material?

MR

 

[Altaic]

Barometric pressure should not make any noticable difference, because the two chambers will be the same pressure. The equal pressure is achieved just before the silicone sealant is applied, when the chambers are open, at equal pressure with the outside world, so matching them should be easy.

I thought the accoustic foam would absorb the inverted wave created by dipole ESLs. It would probably eat up my bass, though, sort of opposing what a baffle accomplishes.

The point of this is to get better sound out of a panel. For not too much more effort and hardly any more cost (maybe $20) arching would be prevented, and max voltage and travel increased. If the volume is increased, then I could put a quazi-enclosure on it, and save some of my bass (maybe... I'm not too sure).

 

[MRehorst]

On a day when the barometric pressure is higher than the day when you sealed the speaker the outer diaphragms will be pushed toward the stators. When playing music, the outer diaphragm will move in step with the driven diaphragm. If the outer diaphragms touch the stators they will buzz. Make sure you have large gaps between the outer diaphragms and the stators. Baromateric pressure changes can create quite a large change in the surface of a diaphragm.

I once considered making a "mirror" out of metalized film by stretching it over a large circular frame, then sealing the chamber behind the film. The result would be a mirror that on some days is convex and other days concave, due to variations is barometric pressure. I even considered making a crude, but very large telescope mirror that way, where the focal length could be changed by changing the air pressure in the chamber behind the film. The mirror so constructed would tend to vibrate with air currents and maybe even sound, but it might provide huge light gathering power (how many amateurs have scopes with 1m mirrors available?) for observing the milky way or looking for comets where you don't need a lot of magnification. The light weight would make the thing easy to handle. I don't know if the shape would be right for a telescope mirror. Maybe I'll try a small one some day. One could most certainly make a dandy solar cooker or light concentrator that way...

The outer diaphragms are very close to the driving diaphragm. That means there is a relatively small air space between them. That means that the driven diaphragm will be moving itself and the air on either side of it plus the two outer diaphragms, which due to the tension on them, will resist such movement. I think the effect will be equivalent to an increase in the moving mass, which should both lower resonance of the driver and maybe also reduce high frequency output (not necessarily a bad thing).

When you build these, build one without the external diaphragms so you can compare the sound from the two types. It should be interesting...

MR

 

[Altaic]

I'll definitely build a normal one and a sealed one. You're totally right about negative chamber pressure. For some reason I was only thinking about positive pressure. One way to get around this is find out the barometric pressure that day, calculate the maximum it could drop and add that much more N2 to each side. of course, that means one would have to balance both sides. A quick and easy to do it would be to loosen the bolts holding the assembly together (the rubber gasket is compressed quite a bit), add the N2 to both sides, and retighten it.

As far as sound-design goes, that would not be good because the sound directionality would change from day to day. You could always have an oiled tube with a snugly fitting ballbearing in it to slowly equalize pressure. As long as the ballbearing responds more slowly than the lowest frequency, bass response wouldn't be affected.

I'll try all these ideas out and report the results. I don't want this to be too complicated, though.

 

[phishead8]

An experimentalist at heart.


Altaic,

I am glad that you are looking for improvements in ESL design. The idea of a full range ESL that is smaller than 4' x 8' sure is tempting.

I don't want to discourage you from building the contraptions that you are mentioning. But, here are some of my experiences and ideas about ESLs (I am not an expert).

1. The construction of ESLs is always more difficult than I expected. I consider myself a relatively careful person, but sometimes the job requires a lot of patience and it is easy to rush a little faster than I should and end up making a mistake. Usually it was cosmetic, thankfully. Either way, a simpler ESL is necessarily simpler to build. My suggestion is to start with the more standard ESL first. If only for practice. If you don't like the sound, I'm sure that someone will be willing to take it off your hands.

2. The infrequent arc is simply not a problem. Unless you are using a low resistance coating (metalized Mylar for example) you should be okay. I suggest that you spend a little bit of time on making an adjustable high voltage supply. When the weather is humid the membrane may arc more. It is easy to simply lower the bias after the fact. This will also allow you to keep the ESL at the loudest music per Watt from the amp.

3. A flat panel behind the ESL will reduce the sound immensely. I haven't tried an acoustic dampening panel, but those things do not work 100%. I concur with MRehorst, allow yourself the ability to try it both ways. I was simply amazed at the difference when I did it.

I am eager to hear about your results.

I am about to begin another pair of ESLs and wouldn't mind some ideas before I get started. Otherwise I'll just stick with my method from last time.

On another note: where did you get your Mylar? I found some .5 mil Mylar in town, but the minimum order turns out to be $75. On the other hand, I'll probably never need more.

Good luck,
Dan

 

[Altaic]

"On another note: where did you get your Mylar? I found some .5 mil Mylar in town, but the minimum order turns out to be $75. On the other hand, I'll probably never need more."

I use this place for most of the parts (frame, mylar, gasket, stator, etc.):
www.mcmaster.com

Notes:
-Raw materials are at the bottom right on the main page.
-When browsing the sections, do not attempt to use the back button to get from the top level back to the main page, as it will autoforward you back to the page you were just at. Do this three times and it will think you are doing a DoS attack on it and lock you out of the site for a while. It's a total pain in the ***, and is my only complaint about the site. Otherwise, their prices are good, they deliver fast, etc.

 

[phishead8]

Rats!

I was hoping that you found a different source. Their .5 mil Mylar only comes 27" wide. My speaker design (and thus materials) is 20" wide. This doesn't leave me much room to wrap around a bicycle tube and piece of wood for the stretching mechanism.

Thanks though, mcmaster-carr is good for everything else.

Get them to send you a hard copy of the catalog. It's a great thing to have around as a DIYer. Sometimes I just flip through it and find all of the stuff that I didn't know existed. Sometimes problems that I was going to try and tackle myself have a ready made solution for extremely cheap. It sure can alleviate headaches.

Good luck on the project.

-Dan

 

[MRehorst]

Is the dielectric strength of nitrogen much higher than that of air (which is mostly nitrogen, anyway)? One of the old commercial ESLs (was it Beveridge?) used a sealed enclosure flooded with a gas with a very long name that DID provide an increased dielectric strength over air. Check the patents- that's where I saw the gas named. I looked up a source for the stuff once- welding supply houses carried it.

I am inclined to agree with phishhead- arcing really isn't problem, but if you're going to chase it, chase it well...

I think that if you make a few small drivers you'll get a good feel for the sorts of mechanical difficulties you may run into assembling, handling, mounting, and wiring ESLs. The mechanical issues are the real challenge . Finding a place to hide the step-up transformers and bias supply(ies) is some of the trickiest stuff you will face.

I wasn't aware that McMaster-Carr stocked so much useful stuff- thanks for the pointer!

MR

 

[Altaic]

I'd agree that arching isn't so much a problem. The thing is, though, that give it 2x or 3x the voltage, and it will do nothing but arch. Put an innert gas in it and it won't, which means you've made the output of a small ESL pretty decent.

I think you're thinking of the gas they use in mig-welding or is it tig-welding? I forget the differences between them. But I think it's really easy to get; my school has tons of it. I'll get the actual name and report back later on, maybe this week.

Would a torroidal mains transformer be better than a normal one for use as a step-up? I'm not going to but a few hundreds of dollars into an audio transformer just yet.

 

[Altaic]

Wait a second, a mains transformer is not going to support HV! Damn it. I'm going to get a neon sign transformer and try that (115V->15kV). It'd do something like 55V->7kV for me. Then I'll bias it up, if I need to.

BTW, instead of biasing, why not just double the 55V beforehand (getting a 1:2 transformer is easy)? Or does it need the bias supply to deal with the capacitance or for safety or something? And how much do supplies which can be used for biasing an ESL usually go for?

 

[MRehorst]

Wait a second, a mains transformer is not going to support HV! Damn it. I'm going to get a neon sign transformer and try that (115V->15kV). It'd do something like 55V->7kV for me. Then I'll bias it up, if I need to.

BTW, instead of biasing, why not just double the 55V beforehand (getting a 1:2 transformer is easy)? Or does it need the bias supply to deal with the capacitance or for safety or something? And how much do supplies which can be used for biasing an ESL usually go for?

I'm confused . Are you talking about using a power transformer in place of an audio transformer? The ESL needs a bias supply or it won't work, period.

The bias supply should supply only a very tiny amount of current (<<1 mA) somewhere between 2kV-6kV or so. You can either use a small transformer to step up line voltage, followed by a standard voltage multiplier circuit, or you can go with a DC-DC converter. Schematics for both types of supplies can be found here: DIY ESL

A DC-DC converter type supply is a bit more expensive unless you can scrounge a module somewhere, but it is much safer to work with than a line operated supply and allows the voltage to be easily adjustable (turn a pot). It is also much smaller, allowing much easier integration (concealment) with the speaker.

I strongly urge you NOT to use a neon sign transformer for the bias supply. They can be very dangerous. Also, DO NOT, under any circumstances, omit the 10M (or more) resistor from the output of the bias supply.

It sounds like you have a very unusual set of priorities- you're concerned about sealing the driver in an inert gas to solve a problem that isn't really a problem, but you don't seem to care about the poor performance you will definitely suffer by using a power transformer (with core material selected for 60 Hz operation) in place of an audio step-up transformer.

I recommend that you go back to first principles and build a few small, simple, test drivers to get your feet wet before you start diving into trying to solve some of truly minor (and once you build a few drivers, you'll see how really minor most of them are) problems with practical ESLs. You'll find there are much bigger fish to fry then occasional arcing.

Once you hear your first test driver working, you will be inspired to build more and bigger drivers, and you will have the experience to know about things like diaphragm tension, bias voltage, insulator thickness, etc. Working on small drivers will allow you to perfect your assembly technique so that building ESL drivers of any size will become something that takes a matter of minutes (literally), instead of hours. You will quickly find that planning and building the support structure for the driver will be a much bigger effort than making the driver.

MR

 

[phishead8]

Altaic,

I remember being in your shoes. Don't worry it will all make sense eventually.

To keep everybody sane, let's use the same nomenclature. The bias is the DC high voltage on the membrane. The step-up transformer is the audio transformer that connects the amp to the stators.

A regular power transformer (one used for making an amp PSU) simply will not work as an audio transformer. MRehorst is right, the power trannie is usually only good around 60hz. They can save lots of money on the design by not making it linear up to 20khz. That is why audio transformers cost so much, they have to do exactly that. The ESL will not be worth the construction if you do not use an audio transformer. Even the cheapest one will sound brilliant in your speakers.

Now, for the bias supply. I haven't had a bad experience with high voltage yet so I kind of disagree with MRehorst. I used a simple "voltage multiplier" to get the desired voltage. You can find schematics for these on the web, but here is the general idea. You use diodes and caps (each rated at twice the input voltage) to multiply an AC voltage. For each diode and cap you get ~1X more voltage of the input. For example 6 diodes and 6 caps will yield 6x the input voltage. But the output will be DC. You don't want to use a large amount of these caps and diodes (I wouldn't use more than 10) so you need the input voltage to be relatively high (~500volts AC). Thus you need yet another transformer. I had to use a combination of 3 transformers in series and parallel to get the desired voltage from wall voltage. Now, the transformer can be as small (hence cheap) as you can possibly find. The DC bias will simply not pull more than 1mA. And, as Mark suggests, do not forget the 20M resistor. I put mine right after the cap/diode chain so as to reduce the current flow if anything went wrong.

If anything in that previous paragraph didn't make sense, I apologize. I realize that I didn't describe it well.

Anyway, good luck!
-Dan

 

[MRehorst]

Let me explain why a variable supply is better then a fixed supply, then comment on ways to make a variable supply.

It is reasonably safe to say that ESLs are pretty low sensitivity speakers compared to your average driver in a box type speaker. You will generally want the ESL to operate at maximum sensitivity so that your amp will be able to drive it to a reasonably loud volume level. This means you will always want to bias the speaker with the maximum voltage you can use.

The maximum useable bias voltage is a function of many factors. Surface perimeter, thickness of the insulators, dielectric strength of any insulating coating you may (or may not) apply to the stators, quality of finish of the stators (sharp edges?), the relative humidity, and diaphragm tension all affect the maximum bias voltage you'll be able to use. Then there is your tolerance for audible hiss and whine produce by corona discharge. Since there are so many variables, it is just about impossible to predict the maximum useable bias voltage for your specific design. That is why having a variable bias supply is a good thing.

I have found that with a really tight diaphragm, and uninsulated perforated sheet metal stators, about 2 kV is the upper limit for bias voltage before you start to hear corona discharge. With thinly insulated stators, about the best you can do is 6kV. Less than tight or lightly supported diaphragms will reduce these voltages considerably. So, a bias supply that is adjustable from around 1kV up to about 6 or 7 kV is ideal as it will allow you to bias just about any ESL for maximum sensitivity. As a point of reference, my Quad ESL 63s use a 5 kV bias supply.

If you have a fixed bias supply, and connect to your speakers and discover that either the diaphragm flaps back and forth or you can hear corona discharge or both, you're going to have to open up the bias supply and modify it for lower voltage. If the speaker works fine but doesn't play loudly, you will want to increase the bias voltage again by opening up the supply and messing with it.

If you have a variable supply, you can start at a low voltage, and raise it until you either corona discharge or the diaphragm starts flapping. Then just back down on the voltage a little and you're good to go. After you have built a few ESLs you'll really come to appreciate the ability to easily adjust the bias voltage.

You can make a variable bias supply a couple of ways. You can go with the transformer and voltage multiplier type circuit and provide multiple taps to get at different voltages, or you can use the same in conjunction with either a rheostat or variac at the power input side of the supply to get true variable output. Small variacs and 10W rheostats are pretty cheap on the surplus market. If you're a good scrounge, this can be a cheap way to go. All of this stuff is relatively large, heavy, and not so easily concealed (read UGLY).

Another way to go is to use a DC-DC converter that can take a safe, low voltage input and convert it up to the required voltage. This method is generally a little more costly than the voltage multiplier technique, but the result is smaller, lighter, and a lot safer to work with. The DC-DC converter supply shown here
www.rehorst.com/esl
fits on a small PCB. I designed it to fit inside a 2" schedule 40 PVC pipe frame that I built to support the ESL. It is completely hidden from view (other than the bias adjust knob). Now if only I could figure out a way to hide the transformers...

Yet another way is to pick up a surplus fly-back transformer and design an oscillator circuit to feed it. A bit of trouble, but probably the cheapest way to go.

Yet another way is to pick up a surplus low current, high voltage supply from a copy machine or printer. These show up on the surplus market quite often and can usually be purchased for around $10-20. They typically have 117VAC input and provide one or two low-current, HV outputs. Some have trim pots that allow adjustment of the output voltage. Cheap, but relatively large. Making fixed supplies adjustable usually has to be done on the HV side by making up a string of HV resistors so they can be tapped to get the voltage needed.

If you're on a budget, you can use a single bias supply for as many speakers as you want to connect to it. Use wire with decent insulation to run the HV around. Test probe wire is usually rated for pretty high voltages. No, it won't affect the sound if it isn't solid silver or 8 nines OFC...

As you can see, there are many ways to get the required HV. The same is true of most of the materials in the ESL. Building an ESL is mainly a test of your resourcefulness- where do you get the stuff, where can you make substitutions, etc. Literally anything you assemble in which a diaphragm is suspended between two conducting stators will make very high quality sound. I have built up working demonstration drivers out of cardboard, plastic wrap and window screen in a matter of minutes. It really is THAT easy.

MR

 

[Altaic]

Cool, thanks for all the info. I've been really busy lately, so I haven't been able to work on my esl (or even browse diyaudio!) for a while. But, I think I now have a good idea of what to use as far as bias is concerned.

About the audio transformer, though... If I have a high bias voltage, then the input voltage shouldn't need to be as high to get a good amount of volume. But if the input is too low, and the bias is then too high, I'm going to get a hiss. If I'm shooting for 8kV across the membrane (this is a reasonable number, no?), then the optimal situation is 2kV bias and 6kV input. But to get 6kV input from 50V or so is going to require a snazzy audio transformer which will put a rather large dent in my pocket. Reversing step-down transformers seems dangerous, because their primary (my secondary) is usually rated for like 2kV. If I hook transformers in a series to get 6kV, these windings will be way out of spec and I'll probably fry my transformer, right?

Correct me if I'm wrong, but I'll either have to drop a lot of cash on audio transformers, or settle with low-volume (not so bad, depending) or hiss (bleh). I'd much rather be wrong/corrected =)

 

[MRehorst]

Connecting transformers in series will not fry them. I have been doing it for years and have yet to toast a transformer, even though I typically drive the speakers with a 200W/ch amp.

8kV is likely too high a bias voltage to shoot for. It is extremely hard to keep such a high voltage from hissing. 5 or 6 kV is really the practical limit, and those sorts of voltages require a lot of effort to keep quiet. You should be able to run >2kV bias without too much difficulty- a simple coat of latex paint on the stators will insulate them well enough to go higher than 2 kV.

You want both the highest voltage audio you can muster, and the highest bias voltage you can use without the speaker hissing. Even at that, the speakers will not come close to the efficiency of a typical dynamic driver system. But unless you like to listen at ear-bleeding levels it should be OK.

ESL sensitivity is a function of the insulator thickness (cheap to adjust), bias voltage (also cheap to adjust), drive voltage (expensive to adjust). If you build a driver and the volume is inadequate, the first thing to try is cranking up the bias voltage. If that doesn't get you to where you want to be, try making the driver thinner by using thinner insulators. If that proves to be too limiting in terms of maximum output, then increase the drive level by either feeding the speaker from a higher powered amp or change/add an extra set of transformers. You can test the double transformer scheme on one of your drivers by using both transformers (you're buying two, for two speakers, right?) on one driver. Then you'll get a good idea of how well that will work for you.

Another cheap way to use a set of inefficient ESLs is to listen to them near field. Put them up at ear level and sit 2 or 3 feet from them. It looks ridiculous, but since the driver operates full range, there are no driver integration issues to worry about. Near field listening almost completely eliminates room effects from the sound (the direct sound from the speaker arrives at your ears much sooner than any reflections from room boundaries), so it is a lot like listening to headphones, except it doesn't sound like the music is coming from the middle of your head.

The most expensive way to go is biamp the system. When the ESL is crossed over at 100-200 Hz, you can make the driver quite thin because it doesn't need to make big excursion for even high output level. That will increase its sensitivity, and since you're now biamping, you'll be able to get the low bass that the ESL lacks.

The nice thing about ESLs is you can build the system up in stages as time and budget allow. Build the ESLs and run them full range with a single transformer. After the magic wears off, and you want higher volume level, add a second set of transformers. Then, when you want more bass, add an active crossover, another amp, and some bass drivers.

Building any ESL system will get you a great sounding pair of drivers right from the beginning, and then you can modify and juice up the system as much as you want. It is the perfect project for DIY maniacs because you can just keep adding on to the original system without having to tear things up or throw away anything you put together.

MR

 

[Altaic]

Ah, but what I was asking about regarding the transformers frying and 8kV was the drive voltage, no? The voltage out of my amp, that is-- shouldn't I maximize this and minimize the bias voltage? Say I typically have 50V out; I'd use a 1:100 transformer to get it up to 5kV, then bias it up more if I need to. I'd need to spend $$$ on that 1:100 transformer, or put them in series, but a cheap transformer won't be rated for 5kV, it'll be rated for 2kV or something.

Here's a rather poor textual rendition of a wave going to an esl, and the factors:
===============================
_......................._........................
..\...................../...\.......................\
....\................./.......\......................| drive
......\............./...........\.................../.....
............................................................\
.............................................................| bias
............................................................/
_________________............<- zero
........................................
........................................
........................................
........................................
......\........../.............\.......
........\....../.................\.....
..........\_/.....................\_
===============================

I'm probably messing it up, so please correct me if I'm wrong.

Edit: Ack! You have to view that erm... with the right font =/

 

[fdegrove]

SEALED ESL

Hi,

Just for info:

Is the dielectric strength of nitrogen much higher than that of air (which is mostly nitrogen, anyway)? One of the old commercial ESLs (was it Beveridge?)

I think you are referring to Dahlquist.





Cheers,

 

[MRehorst]

You do not have to worry about frying the transformers. Maximize both drive voltage and bias voltage.

I have used one set of 25W @30 Hz rated tansformers for four generations of speakers with bias as high as 5kV and always driven by a 200 W/ch amplifier. Frying transformers simply is not a problem. The speakers will arc long before the transformer cooks, and hopefully, you'll have enough sense to turn the volume down a bit when the speaker starts arcing (they should be playing painfully loud at that point anyway).

MR

 

[SY]

The nitrogen thing puzzles me, too. If you're going to go to the trouble of sealing up a unit with gas, use something with significantly higher dielectric strength than air. Dayton-Wright, many years ago, used sulfur hexafluoride (SF6), which is inert, nontoxic, has a very high dielectric strength (you can use higher voltages), leaks more slowly than nitrogen/air, and gives better mass-loading to the diaphragm because of its high density. It's a bit expensive, but that would be lost in the noise of all the other things you're buying.

What's cool is that if you huff the leftover gas, you get the exact opposite effect from huffing helium- your voice gets really, really deep. In some experiments I was doing, we were breathing mixtures of SF6 and oxygen (don't want to suffocate!), and we found that we could do terrific imitations of Paul Robeson.