posted 19 March 2002 04:48            

stereoization.kym

The Sound includes a pink noise source (which is mono!) so you can try different delay settings, up to a max of around 20 ms. Using this noise source, if you keep the attenuation of the higher frequencies below about -6 dB and turn off the low frequency EQ (-30 dB) then you can just barely detect the slightest coloration of the sound as you alter the delay.

I find that 20 ms is really a good setting as the coloration occurs mainly below those enhanced frequencies. I suspect, but I haven't examined the situation that closely yet, that 20 ms delay is causing a comb filtering with about a 50 Hz base frequency. Hence the comb teeth are probably very closely spaced at the higher frequencies.

20 to 50 ms is about the longest tolerable delay before you start to recognize distinct echoes. At 50 ms you can detect slap-backs, but at 20 ms these tend to blur. If you pass equal amounts of energy at all frequencies with a 20 ms delay, then you can detect flams on drum hits.

But in this case I am attenuating the low frequencies and passing only small amounts (< about half) of the higher frequencies. Higher frequency here means above 500 Hz. Low frequency here means around 250 Hz. By adding only small amounts of HF, you minimize the comb filtering artifacts.

Coupled with the 20 ms delay on the right channel, it tends to spread the stereo image somewhat. This is not an in-your-face kind of effect, and depending on your EQ settings it can be subtle but delicious.

The Haas effect, or precedence effect, states that a sound appears to arrive from whichever channel it first appears in, regardless of its amplitude to a large extent. Here I am confounding the effect somewhat by passing only the higher frequencies as delayed. [er... rather, because of the Haas effect and the direct feed, I am not upsetting the stereo imaging of the sound by adding a little delayed right HF.]

The masking occurs because a direct sound generally appears early and louder than the uncorrelated reverberation and ambience. Hence while the direct sound is strong, you can't notice the higher frequencies added and delayed. (In fact I'm not delaying them in the left channel here at all... but the right channel is delayed, which allows us to stereoize a mono source.)

Finally, the low frequencies which carry drum strikes and bass generally are direct paths and very strong, compared to the higher harmonics which are also direct but very short, followed by uncorrelated ambience and reverb from them.

The end result is to trick the mind into perceiving a soundstage with more depth, greater clarity for the instruments in front, and an increased sense of sourrounding ambience.

Note that I don't decode anything into M/S channels here. It isn't needed. Just a selectively delayed EQ of higher frequencies.

Enjoy! (I sure am!!)

- DM

[Finally, it just occured to me that there is probably no need to feed EQ HF to the left channel at all... Try ripping it out and see what you think. Ripping it out of the left channel will restore the original EQ balance of the music, since both left and right now carry the original program material, and the right channel has an attenuated and delayed copy of the HF added. The delay is long enough that most of the program material will not be upset by the small amount of extra HF....

By the same reasoning, try adding the same EQ of the left channal back in with a delay, but make the delay about 5 ms longer than the right channel delay, e.g., 25 ms Left and 20 ms Right. The result is even more delicious! (I find that attenuating the HF bands by -4 dB works slightly better than -6 dB, but my hearing is not so good either...]

posted 20 March 2002 02:18            

The comb ripple is present in both amplitude and phase response. It is largely reminiscent of a typical room response with about a 3 dB p-p ripple amplitude. But unlike a room response there is no strong resonance in the 200 Hz region.

The system I tested had a 15 ms delay in the left channel and a 20 ms delay in the right channel. When you examine both channels together on a spectrum analyzer you find that they weave back and forth in their ripples and produce a complex multiband EQ on the left and right channels that have few coincidences.

So it is apparent now why this sounds so much like listening in a real room -- it approximates a room-like ambience and gives the sounds a more 3-D feeling. It does a better job of shuffling frequency response between left and right channels than any available graphic EQ. There are so many more bands.

As predicted, these ripples have periods of about 100 Hz and 133 Hz respectively (= 2/delay). No EQ I have ever seen can duplicate this kind of behavior. There are just too many interweavings.

posted 22 March 2002 02:05            

stereoization.kym

I have heard mention of bass enhancement by "pulse stretching" at low frequencies. Whether this is how it is done or not, who knows, but by adding in delayed amounts of L and R back to their respective channels (no cross mixing!) and by attenuating the highs above 500 Hz by about 4 dB, but keeping the bass at 0 dB for fold-in, this effectively "stretches" the low frequency waves and enhances the bass dramatically. The delays used here are 15 ms and 19 ms on L and R respectively. They are changeable on the VCS.

Add to this, some stereo spreading by first encoding the signal to Middle/Side, and applying gain to the Side before reconstructing the Left and Right channels ahead of the 3-D treatment -- drastically deepens the stereo image without upsetting the original placement of various instruments. A Side gain of about 4 dB is quite dramatic. An attenuation is applied to the original stereo signal to prevent it growing too loud with this extra Side gain.

By comparison, normal stereo in headphones is like looking at a 2-D picture. Following all this processing I normally listen through M/S cross encoded headphones to relieve the differential ear pressure. None of this processing interferes with that encoding. That M/S encoding for headphones makes use of a short 600 - 800 usec delay to provide psychoacoustic cross mixing of attenuated Side signal. That way both ears hear some of every signal, but the signal is still strongest in the correct L/R channel. [The Haas effect is used productively here too to retain correct side placement of sounds...]

The bass enhancement with this not-so-subtle processing is astonishing. I can almost feel my whole body vibrating in response to the bass beats. It is an illusion, but it is a good one!

- DM