In practice, it took me less than a minute to figure out what Crystalizer really was. This technology is similar to what we have seen in Intel with its HDA, in the bundled Intel Audio Studio.

Indeed, a record is remastered by a well-known mastering plug-in called "multiband compressor". Just to make sure I was 100% correct, I compared Crystalizer with three mastering multiband compressors: Waves LinMB, iZotope Ozone, Steinberg MultiBand Compressor. These plugins changed the audio character similar to Crystalizer.

Our measurements showed that besides the multiband compression, the signal level is raised approximately by 3 dB. So that any quiet records would seem subjectively better even without the compressor.

24-bit Crystalizer technology has a right to life, but the way this technology is announced with a portion of wishful thinking is disappointing. In reality, Crystalizer does not expand, but narrows down the dynamic range. It really uses 24 bits, but this is done only to avoid the rounding error accumulation (it's normal practice, no modern DSP works in the same resolution as the original data).

"There is one very important issue, which is not exactly advertised by the manufacturer - unlike the audio processors of the previous generation, X-Fi allows to disable resampling as well as other DSP effects! In this case you lose the unique feature to listen to several parallel signals with different sampling rates and apply various effects and improvements. But if you love maximum audio quality, you need this feature the least of all."

ASIO 2.0 is supported with < 2mSec latency. 0% CPU overhead is required to pass audio to/from the host to hardware.

The situation with MP3 and WAV is similar, but there is some difference. Mastering studios use very expensive equipment that analyzes the original file: not in the real time, but running a tad forward, or even at several passes; and the calculation precision is floating point 64 bit instead of 24 bit. Then, mastering compression parameters must be laboriously adjusted individually for each record on high-quality reference monitors of the middle and far field. To delegate a similar function to real time automation and then speak of the advantage over the studio results is arrogant at best. The final mastering stage includes a noise shaping step, which makes a 16-bit record sound like a 20-bit one. So the only thing left is to transfer the record bit-by-bit to the DAC input. Any interference into this process damages audio quality. Anyway, an MP3 file cannot possibly be better than the original record, multiband compression may distort the tone balance and result in the redistribution of signal energetics, which will lead to the overload at high or low frequencies.

It's quite another matter, if we speak of mediocre consumer acoustics with a limited dynamic range and defective frequency response, so that even the most primitive timbre adjuster makes a record subjectively much more comfortable to listen to. In this case, an additional band compression of high and low frequencies facilitates the record perception through given speakers, especially combined with moderately compressed original record, with less aggressiveness (peak power/average value). In this case the resulting audio processing distortions will not exceed the distortions in the playback channel.

That was the situation in practice. Some records sounded more coherent on inexpensive active speakers, though with more aggression. In case of high quality acoustics (studio monitors and Hi-Fi speakers of the mid class), I had no problems picking several MP3 files as well as original CDs, which quality was degraded by Crystalizer. I heard overloads, or the sound got too aggressive, so that I was getting tired of listening very fast.