• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

So I decided to try and figure out what the X-fi Crystalizer actually is

Phos

Member
Joined
May 27, 2021
Messages
60
Likes
39
The X-fi is kind of a funny sound card, because it was made with the looming threat of motherboard integrated audio being "good enough" for most consumers, so I think Creative (Cre^tive?) felt it needed to be all things to all people. As a result, this card has both hardware acceleration for spacial audio in games (just in time for Vista to remove support for most of it) and true ASIO support.

But this is about the "Crystalizer". The recent MQA kerfuffle got me curious to see if I could figure out what it was actually doing. I think it was sometimes described as the "24-bit crystalizer", and as I recall was described in the marketing as making compressed recordings sounds like masters without getting into any specifics (remind you of anything?). At the time I didn't bother much with it, it kind of just sounded like it was bumping up the volume a bit, which makes level matched comparisons tricky.

The controls for the crystalizer consist of a checkbox that turns it on and a slider that goes from zero to 100. For some reason zero doesn't actually seem to be zero, because there's still an audible change when you turn it on and off even when set to zero. I can't tell if Creative was trying to obfuscate what the crystalizer does, because it doesn't work on the S/PDIF out, only on the "speakers" out (Which is also the device for headphones out, the front panel isn't its own output device), BUT you can still get an all digital loopback because basically all soundblasters come with an input device called "What U Hear", which is just the speaker output as an input... and you can select to play the speakers' stereo mix over S/PDIF as well.

So the first test was "what does it do to sine waves?", which also revealed a probable reason for why it doesn't impact the digital output, because if windows is set to max volume the answer is "add clipping" if the sine wave was also at 0 db. So it is increasing volume somewhat. Probably not a problem for analog, unless, you know, you want line level output into an amp, as 100% volume means 2V (Which was really fun when a previous motherboard started causing a hard crash that ended with white noise at max volume, because for, let's just call it multimedia, I just have a PC360 plugged into my front panel. It wasn't the sound card, it even with it completely removed). It did reveal that

So then I decided to take advantage of the aforementioned "What U Hear" option, and had audacity record some music being played through the crystalizer, and compared it to the source file, this is (some of) "Electric Eye", crystalizer is on the top, 320 kbps MP3 source below:
1622094419654.png

Was almost more surprised to see Electric Eye looking so loudness war'ed, I got it during the time when my go-to way to get new music was to buy an old CD of it off Amazon second hand for like $2, IIRC that's a rip from the 1982 original.

This process only proceeds at real time so I didn't go through the whole song. SO what we have here is a dynamic range DEcompressor. The audacity recording doesn't include any clipping that I heard, so it seems the "What U Hear" input skips adding any digital gain to the signal. Not gonna lie, without the extra gain it seems to have improved it somewhat, but it's not a perfect test because I forgot to turn off my EQ profile in foobar.

So I seem to have answered my question, the crystalizer does add a bit of digital gain but that seems to be to avoid the decompressor's output being quieter than normal. But regardless of if it's any good or not (particularly for any tracks that have been loudness-war'ed), it's not very convenient to use, it only works on some outputs if you know how and seems to have been meant to be used in a set it and forget it manner (meaning it's kind of buried in the creative control panel) and I don't think you want it on well mastered tracks.
 
OP
P

Phos

Member
Joined
May 27, 2021
Messages
60
Likes
39
So I discovered that trying to play 88200 Hz media over the X-Fi's digital outputs using foobar in WASAPI event mode bluescreens my PC. It'll play the file fine in WASAPI over the analog out, but for some reason it just doesn't have support for 88200 Hz for S/PDIF.

Since it's an all digital loopback I figure RMAA is probably be perfectly adequate, aaaand the plot thickens, this is with the Crystalizer off (white) vs Crystalizer on set to 65% (green), No audio was converted to analog for this test:
1622248353812.png

Not quite the Harman target curve, more likely Creative just assumed whatever cheap speakers people would be using would be mid dominated.

So what's actually pretty interesting is what the level adjusting screen looks like with crystalizer on:
1622249272199.png

Note that the colors are different channels rather than different tests for this graph.
No idea why there would be a channel imbalance like that, if I compare left and right in the actual tests one line occludes the other. Regarding the distortion spikes, it isn't clipping, the level is correct, and the actual THD results look like this:
1622249443175.png

(back to colors for on/off) More harmonic distortion with it on (still below audibility), but way lower than the test tone, I think what's going on here is the abrupt start of the tone (for the level matching pre-test) is being detected as a transient. But wait, those extra spikes aren't harmonics of anything, not sure what's going on there.

1622252140698.png

Spoilers I'm not writing this post entirely in order and RMAA crashed so this is with crystalizer set to 55%, but the not-harmonics are also here.

1622250001941.png

We can see here that the level of the EQ-ing changes with the % that the crystalizer is set to. For some reason the crossover points are also different, which isn't quite what I would expect. I don't get the ultrasonic EQ shelf it's doing, are they trying to reduce the audibility of hypothetical aliasing? The tests were all done at 24/96K.

Just for fun, I ran it once with CMSS-3D turned on, you can think of it as emulating a pair of virtual speakers playing to a pair of virtual ears. It's for games:
1622251079193.png

Did they emulate a comb filter? The large amount of crosstalk is to be expected and doesn't represent a problem with the algorithm (It's almost like this wasn't what this test was designed for!) CMSS-3D works way better on games with old style hardware audio and EAX support, you get true binaural out of it. I just noticed that the crystalizer somehow reduces crosstalk at low frequencies even further even though it's entirely digital?

I tried to compare CMSS-3D to "windows sonic for headphones", but didn't show anything remarkable in RMAA. Maybe it's smart enough to only turn on when something more complex than stereo is playing?
 
Top Bottom