Why does AMP sound so smooth and clean?
───────────────────────────────────────
The major advantage of AMP is the direct use of the EMU8000 synthesizer that
performs the following tasks in hardware (that means with no CPU load):
1. high quality pitch-shifting using a patented dynamic four-point
interpolating filter
This allows the removal of signal distortion that is so typical for all
software-mixing players. Their interpolators (if any) are mostly two-point
linear or rarely three-point quadratic (or they are unknown - hidden behind
magic and meaningless names like FFT, FOI, IDO, 32-bit, spline etc.). But
those basic mathematical algorithms are not suitable for signal processing,
therefore such interpolators can't remove the unwanted artifacts while
preserving the high frequencies contained in the original waveform.
<< NOTE: For those of you being unfamiliar with digital signal processing,
here is a brief explanation:
Let's take an example. You have an instrument sampled at 20kHz when playing
the C4 note. But you need to play e.g. the C3 note at 40kHz output rate.
To do it, you must increase the number of sample points (i.e. decrease
the pitch) by factor of four _without_ changing the information content.
If you obtain the three new points by repeating the old sample value, then
the resulting frequency spectrum will include new high-frequency replicas
(mirrors) of the original signal's spectrum. In our example, the original
spectrum 0..10kHz would be shifted to 0..5kHz and also mirrored to ranges
5..10kHz, 10..15kHz and 15..20kHz. These replicas would add an ugly ringing
distortion to the output sound. To remove them, you need a sharp-cutoff
low-pass filter that won't damage the signal that is passed through.
The result will be the "right" new samples smoothly inserted between old.
This kind of digital filter is often unfortunately called an "interpolator"
even if it does not perform "interpolation" in a common mathematical sense
(such as linear or polynomial interpolation between known points).
It is important to realize that a well-designed filter response is more
crucial then just a number of input sample points used for interpolation.
>>
2. mixing at 44.1 kHz frequency allows output frequencies up to 22 kHz
You would need a very fast Pentium machine to get just a simply interpolated
sound in 30 channels at the 44kHz mixing rate with a software-mixing player.
And it would eat most of the CPU processing power.
With AMP, all you need is a Sound Blaster AWE32 compatible sound card with
sufficient sample RAM and any 386 or better computer.
3. starting/terminating the notes and changing the volume without any clicks
(using the EMU8000's envelope engine)
The software-mixing players usually can't afford such smooth volume ramping.
4. customizable reverb and chorus effects
Without effects, the sound is dry and flat.
Any good effect processing requires many additional complex computations.
5. smooth panning in 256 steps
<< NOTE: I've heard a few people saying that software mixing performed by
their interpolating routines sounds better than hardware mixing on the AWE
card. That's a big nonsense from both theory and experience. High quality
pitch shifting and mixing equivalent to that performed by EMU8000 (four-point
resampling in 30 voices at 44.1 kHz) requires about 12 million multiplications
per second! << (4+4+1) * 30 * 44100 = 11,907,000 >>
This means that the sole multiplication would consume all cycles on a P120
machine. But just multiplying is not enough. We have to do other things as
well. And what's even worse, we still don't have any reverb and chorus.
The MMX technology can be 3-10 times faster, though - that may be a real
chance for software-mixing players. Time will show.
(If anyone wants to see the details on how the EMU8000 works and how complex
the computations are, I recommend checking the US patent # 5,111,727 -
available also for free on Internet, e.g. from the Patent Server operated
by IBM).
>>