Is OPL3 the same synth as Creative Music Synth [220]

[Green Xenon]

Both channels are identical. If I invert the left, and mix it with the right, it cancels out completely.

Sorry man but you're just wrong, there's no phase errors/features here. I don't know what you did to it in your first zip.... Uhm... I think you need a bit more experience in the studio before this kind of analysis.

Impossible. I've tried playing OPL3 audio through my Karaoke Voice Canceller [], the effects are the same as if I invert the left channel and mix it with the right using audio software [such as Adobe Audition or Wavelab]. The audio is more pleasant with than without the voice-canceller.

The Karaoke Voice Canceller is a hardware analog device outside of the PC that does pretty much the same thing -- invert one stereo channel and combine it with the other channel to get a mono without what was identical in the original L and R channels.

I can use audio software or analog audio hardware. The effect on the OPL3 audio is the same. I've done this with several different audio softwares [including Wavelab and Adobe Audition], as well as hardware analog karaoke voice-cancellers.

Are you sure you didn't accidently do something that would cause them to completely cancel out?

This is the voice-canceller I currently have:

http://cgi.ebay.com/Optimus-Karaoke-Voice-Canceller-with-plug-audio-wires_W0QQitemZ380022294749QQihZ025QQcategoryZ64604QQrdZ1QQssPageNameZWD1VQQcmdZViewItemQQ_trksidZp1638Q2em118Q2el1247

[stryd_one]

Edited:

I did this with a new audio editor and screwed up - they aren't the same on left and right at all...

End edit

The audio is more pleasant with than without the voice-canceller.

If you think that second example from the first zip sounds "pleasant", then I think we're done here. It sounds like %$@#, it's almost more noise than signal.

Forget your voice canceller, it's a proprietary device and therefore a mystery. Are you sure it's analog? an analog device doing what you suggest would have no knobs, so I wonder what setting you've used and what it really does. No need to introduce unknown variables. Just work with the waveforms in wavelab or whatever, no VST plugins or anything like that, just flip one channel and then mix them together.

There's a lot of background noise on that file, was it recorded digitaly or did you pass it out of the soundcard and back in again? What else was in the signal chain? The reason I ask this, is that the only reason that the sound does not cancel out completely, is a slight phase shift (delay) of the entire waveform on one channel.... The right channel is just a little bit earlier than the left. I'll let the screencaps say it. The second one is zoomed in so you can see the level difference and it's relation to the mixed file.

What you're hearing after the flip and mix, is the difference between the two channels. In the lower frequency parts of the waveform, the levels change slower, and because the channels are only slightly out of phase, there is a good chance that they will sample at the same level, so they will cancel out completely. The higher frequency portions of the sample, change amplitude faster, so there is a greater chance of them sampling at a slightly different level, so they are left behind. Essentially, it's a highpass filter.

Maybe you just like the sound of PC noise and high frequencies ;)

peak1.JPG

peak2.JPG

[Green Xenon]

Open the file you just posted in wavelab and look at the waveforms for left and right. They're identical.

There not the same, you've got to phase-invert one channel and convert to mono to see the difference. Tomorrow, I'll send you the stereo file with the left channel's phase inverted. Then you can convert the file to mono and see if anything shows up.

If you think that second example from the first zip sounds "pleasant", then I think we're done here. It sounds like %$@#.

Not everyone has the same audio preferences. What one hates, another might love. For example, I've read many posts on Usenet of people saying bad things about OPL3. They call SB16's FM synth "cheesy" and "tinny". Of course, as a intense fan of OPL3, I strongly disagree with those posters.

Forget your voice canceller, it's a proprietary device and therefore a mystery. Are you sure it's analog? an analog device doing what you suggest would have no knobs, so I wonder what setting you've used and what it really does. No need to introduce unknown variables. Just work with the waveforms in wavelab or whatever, no VST plugins or anything like that, just flip one channel and then mix them together.

With or w/out these "unknown variables", phase-inverting one stereo channel and then combining it w/ the other channel [to convert to mono] seems to have the identical results I previously described.

[NV]

I took your polypadc5.wav file, inverted the left channel then mixed it with the right in ProTools, and I did hear a difference. However, I fail to hear the "heavenly" qualities you are referring to (which is admittedly subjective). The inverted audio had a "sharper" quality to it, although the sound is certainly not "equally loud."  In order to achieve normalized volume between the original and inverted files I needed to boost the inverted audio by about +30db, at which point the inverted synth sample was less in volume than the noise inherent in the file. Beyond the noise, to me it sounded quite similar to the effect of running the original file through a highpass filter and some minor equalization to achieve more of a reed-like quality by eliminating some of the low and mid frequency content.

All of this aside, I fail to understand what exactly you are attempting to find. Why the synth would be designed this way? It may have nothing to do with the OPL3 itself, but instead with the particular chipsets which you have been using. Even if your chipset is working perfectly, and all of the OPL3's were designed with this phase differential in mind, I don't think there is necessarily any holy grail answer as to why this could be. There are far too many variables in this equation for any sort of definitive answer save from Yamaha themselves, and I can certainly understand Creative's willfull ignorance when you question them about a minor characteristic of an obsolete technology they utilized nearly two decades ago.

I do not yet have a working OPL3 synth which I could use to test and see if this effect is a universal characteristic of the OPL3 chip, but frankly I don't know if it would matter. There have been many pieces of stereo equipment designed in the past which contained minor phase discrepancies, particularly in less "professional" level equipment, a niche into which the OPL3 would fit. I don't believe that an inexpensive chip of a consumer-level computer soundcard manufactured in the early 90's ever had any intention of achieving audio perfection.

[stryd_one]

Please check my edit :)

[stryd_one]

There not the same, you've got to phase-invert one channel and convert to mono to see the difference.

I'm not a n00b, I'm just working with a new audio editor, and copied the wrong channel when i paste+mixed it.

It's pretty clear that they *should* be the same, there's certainly no change of phase within either channel as you suggested, just a linear delay of the left channel.

[stryd_one]

Not everyone has the same audio preferences. What one hates, another might love. For example, I've read many posts on Usenet of people saying bad things about OPL3. They call SB16's FM synth "cheesy" and "tinny". Of course, as a intense fan of OPL3, I strongly disagree with those posters.

LOL there's a big difference between disliking a synths tone, and disliking a mass of noise. The 'voice-cancelled' version is about 40% noise. That's not a matter of taste, that's just %$@#!

[NV]

I just read through Stryd_one's recent edit and I must agree with his findings. The sonic characteristics I noticed in the inverted audio file were very much akin to the character of the phenomenon stryd describes. This could be indicative of a minor shortcoming in your audio chain, something hardly noticeable in standard use but detectable with phase inversion.

You mentioned that you had tried the same experiment with other people's soundcards as well as your own. Did you utilize the same signal chain? While there may be a miniscule phase shift inherent to the design of the OPL3 (or perhaps of the SoundBlaster 16 cards you have been testing), it may also be a kink in your audio chain. Should you find the same results using different signal chains or rule out the chance of a phase shift in your own setup then that may be indicative of the design of the OPL3, although it may also be the design of the specific cards you have been testing rather than the chip itself. A sure fire way to test this would be to generate a similar sample out of an OPL3/YAC512 synth that is independent of the SoundBlaster architecture, such as the MBFM. Beyond that, there are still many variables which could quite easily contribute to this which may rest outside of the OPL3's design.

[Green Xenon]

I took your polypadc5.wav file, inverted the left channel then mixed it with the right in ProTools, and I did hear a difference. However, I fail to hear the "heavenly" qualities you are referring to (which is admittedly subjective). The inverted audio had a "sharper" quality to it, although the sound is certainly not "equally loud."  In order to achieve normalized volume between the original and inverted files I needed to boost the inverted audio by about +30db, at which point the inverted synth sample was less in volume than the noise inherent in the file. Beyond the noise, to me it sounded quite similar to the effect of running the original file through a highpass filter and some minor equalization to achieve more of a reed-like quality by eliminating some of the low and mid frequency content.

All of this aside, I fail to understand what exactly you are attempting to find. Why the synth would be designed this way? It may have nothing to do with the OPL3 itself, but instead with the particular chipsets which you have been using. Even if your chipset is working perfectly, and all of the OPL3's were designed with this phase differential in mind, I don't think there is necessarily any holy grail answer as to why this could be. There are far too many variables in this equation for any sort of definitive answer save from Yamaha themselves, and I can certainly understand Creative's willfull ignorance when you question them about a minor characteristic of an obsolete technology they utilized nearly two decades ago.

I do not yet have a working OPL3 synth which I could use to test and see if this effect is a universal characteristic of the OPL3 chip, but frankly I don't know if it would matter. There have been many pieces of stereo equipment designed in the past which contained minor phase discrepancies, particularly in less "professional" level equipment, a niche into which the OPL3 would fit. I don't believe that an inexpensive chip of a consumer-level computer soundcard manufactured in the early 90's ever had any intention of achieving audio perfection.

Now I'm getting a bit concerned that this effect I described is a result of Creative Technology intentionally doing this. While I like the effect, I'd like to know more about it and CT ain't gonna tell me @%@$! about it.

CT are so cruel!

Anyways, I get the feeling that whoever designed the chip did intentionally design it that way. This is because, they were probably trying to place sounds with "vocal" qualities in the center [i.e. in-phase for both L and R] and sounds with "chorus-like" qualities phased-differently in L and R. I think they reason they did this is because they wanted to copy what the musical industry does. Normally, in most pop-music, the bass, lead vocals, kicks, and drums are recorded identically in both the L and R channels [which explains why most voice-cancellers decrease the bass], while the piano, pads, chorus, and guitars are recorded non-identically in the L and R channels. Once again, this is just my guess but it seems to me that this is what the chip-designer intended.

[Green Xenon]

I just read through Stryd_one's recent edit and I must agree with his findings. The sonic characteristics I noticed in the inverted audio file were very much akin to the character of the phenomenon stryd describes. This could be indicative of a minor shortcoming in your audio chain, something hardly noticeable in standard use but detectable with phase inversion.

You mentioned that you had tried the same experiment with other people's soundcards as well as your own. Did you utilize the same signal chain? While there may be a miniscule phase shift inherent to the design of the OPL3 (or perhaps of the SoundBlaster 16 cards you have been testing), it may also be a kink in your audio chain. Should you find the same results using different signal chains or rule out the chance of a phase shift in your own setup then that may be indicative of the design of the OPL3, although it may also be the design of the specific cards you have been testing rather than the chip itself. A sure fire way to test this would be to generate a similar sample out of an OPL3/YAC512 synth that is independent of the SoundBlaster architecture, such as the MBFM. Beyond that, there are still many variables which could quite easily contribute to this which may rest outside of the OPL3's design.

What do you mean by "audio chain" or "signal chain"?

[NV]

Anyways, I get the feeling that whoever designed the chip did intentionally design it that way. This is because, they were probably trying to place sounds with "vocal" qualities in the center [i.e. in-phase for both L and R] and sounds with "chorus-like" qualities phased-differently in L and R. I think they reason they did this is because they wanted to copy what the musical industry does. Normally, in most pop-music, the bass, lead vocals, kicks, and drums are recorded identically in both the L and R channels [which explains why most voice-cancellers decrease the bass], while the piano, pads, chorus, and guitars are recorded non-identically in the L and R channels. Once again, this is just my guess but it seems to me that this is what the chip-designer intended.

I can understand your rationale, but I don't believe that Creative or Yamaha did this intentionally. If you read through Stryd_one's edit of his recent post, you will see that the phase issues you are discussing are not a product of intentional phase differential so much as they are a product of a very minor phase shift (delay) between the left and right audio channels. This is why we mentioned the potential issues with your signal chain.

What do you mean by "audio chain" or "signal chain"?

A signal chain is the path that an audio signal (voltage) takes as it passes from the source to its destination. For example, if you are playing a guitar through an amp and cabinet, the signal chain would be as follows:

Guitar -> amplifier -> cabinet/speaker

Or if you were recording a hardware synthesizer through a pre-amp and into your computer sequencer, the path would be:

Synthesizer -> DI box/pre-amp -> A/D converter -> audio interface/sound card -> sequencer/hard drive

When working with a signal chain there can often be a small problem at a given point of the chain. Say for example, when you are playing your guitar you suddenly notice a great deal of unfamiliar noise. This could be any number of things along the chain - your guitar pickups could be receiving interference, your amplifier tubes could have microphonic issues, or your speaker wiring could be faulty, to name only a few. So in order to figure out where the problem is you have to do some troubleshooting through the possible variables, eventually resulting in discovering and fixing the problem.

When Stryd and I ask about your signal chain we are wondering what chain you used to record your samples, or the times when you have noticed the phase issues. Since the phenomenon appears to be the result of a minor phase shift, there are a number of possibilities along your signal chain that may contribute to this. Or it may simply be the brand of soundcard, such as the SoundBlaster 16. If there was a common variable between all of the times you have witnessed this phase differential (always using the same speakers, always using the same brand of soundcard, always using the same audio interface, etc) then you may have found the source of all this.

Or it could be the OPL3 chip itself, however the only way to test this would be by utilizing an OPL3 chip independent from the other variables you have encountered, which would mean building the MBFM synthesizer you were asking about to begin with.

[Green Xenon]

My signal chain is all within the sound card itself. No external devices.

[Green Xenon]

Edited:

I did this with a new audio editor and screwed up - they aren't the same on left and right at all...

End edit

If you think that second example from the first zip sounds "pleasant", then I think we're done here. It sounds like %$@#, it's almost more noise than signal.

Forget your voice canceller, it's a proprietary device and therefore a mystery. Are you sure it's analog? an analog device doing what you suggest would have no knobs, so I wonder what setting you've used and what it really does. No need to introduce unknown variables. Just work with the waveforms in wavelab or whatever, no VST plugins or anything like that, just flip one channel and then mix them together.

There's a lot of background noise on that file, was it recorded digitaly or did you pass it out of the soundcard and back in again? What else was in the signal chain? The reason I ask this, is that the only reason that the sound does not cancel out completely, is a slight phase shift (delay) of the entire waveform on one channel.... The right channel is just a little bit earlier than the left. I'll let the screencaps say it. The second one is zoomed in so you can see the level difference and it's relation to the mixed file.

What you're hearing after the flip and mix, is the difference between the two channels. In the lower frequency parts of the waveform, the levels change slower, and because the channels are only slightly out of phase, there is a good chance that they will sample at the same level, so they will cancel out completely. The higher frequency portions of the sample, change amplitude faster, so there is a greater chance of them sampling at a slightly different level, so they are left behind. Essentially, it's a highpass filter.

Maybe you just like the sound of PC noise and high frequencies ;)

The signal chain was within the sound card itself. I did not use any external devices. You are right in that I prefer treble over bass. It also makes sense that it's easier for higher frequencies to lose phase more easily than lower frequencies. However, I think this is only the small [and very small], part of the issue. The waveshapes are actually very difference in polypadC5A vs. polypadC5B. This regardless of how much you try to make their frequencies identical. The shapes are different, hence the audios will have a different quality.

PolypadC5A seems to have the sine-wave qualities of a lead vocal singer. PolypadC5B seems to have the sawtooth-wave qualities of the chorus. Sorry, this is a bad analogy, but it's the best I could express.

This is very similar to pop music. The lead-vocal is usually phased identically in L and R while the chorus is usually phased-differently in L & R. That is what leads me to suspect that whoever designed the chip made this effect on purpose -- even if, in fact, it was not intentionally. It just seems so similar to pop music phasing that one would think the designer wanted to copy pop music in doing this effect.

That's just how I perceive it.

[stryd_one]

The waveshapes are actually very difference in polypadC5A vs. polypadC5B.

Of course they are, the 5B is highpass filtered by the phase difference and inversion and mixing.

But the left and right channels on the original file, are essentially identical - just one of them is late.

You could actually get the same file as 5B, by highpass filtering the original - if the filter response was just right.

This is very similar to pop music. The lead-vocal is usually phased identically in L and R while the chorus is usually phased-differently in L & R. That is what leads me to suspect that whoever designed the chip made this effect on purpose -- even if, in fact, it was not intentionally. It just seems so similar to pop music phasing that one would think the designer wanted to copy pop music in doing this effect.

Nup, this is an entirely different effect. Such an effect you speak of is brought about by phase inversion, and there's none of that in the OPL output.

Here's a project for you - try lining the sound out of that OPL card, and recording it at 96khz. Just for your own homework, but I think you'll find the effect interesting :)

[Green Xenon]

Of course they are, the 5B is highpass filtered by the phase difference and inversion and mixing.

But the left and right channels on the original file, are essentially identical - just one of them is late.

You could actually get the same file as 5B, by highpass filtering the original - if the filter response was just right.

I doubt I would get the same effect by highpass filtering alone. I've done something similar before, using the Adobe Audition software's graphic equalizer to eliminate the bass and amplify the treble. It didn't work at all. The audio still had the quality of a lead vocal singer, and not the quality of the chorus.

So there's more to it than just frequency.

This is very similar to pop music. The lead-vocal is usually phased identically in L and R while the chorus is usually phased-differently in L & R. That is what leads me to suspect that whoever designed the chip made this effect on purpose -- even if, in fact, it was not intentionally. It just seems so similar to pop music phasing that one would think the designer wanted to copy pop music in doing this effect.

Nup, this is an entirely different effect. Such an effect you speak of is brought about by phase inversion, and there's none of that in the OPL output.

Here's a project for you - try lining the sound out of that OPL card, and recording it at 96khz. Just for your own homework, but I think you'll find the effect interesting :)

If OPL3 is a stereo synthesizer, wouldn't it make sense that there will be significant difference in the right signal vs. the left signal?

[stryd_one]

I doubt I would get the same effect by highpass filtering alone.

That's all it is (I explained how before)

I've done something similar before, using the Adobe Audition software's graphic equalizer to eliminate the bass and amplify the treble. It didn't work at all.

Like I said, the filter response would have to be just right.

If OPL3 is a stereo synthesizer, wouldn't it make sense that there will be significant difference in the right signal vs. the left signal?

That depends on the patch.

Try this:

Open the original stereo file

Upsample it to 88200 (twice the rate). Use the best interpolation available in your editor.

Add ONE sample to the beginning of the right channel

flip one channel and mix them together

Compare that with the same thing without the one sample added. Are you getting it yet? ;)

[seppoman]

Sadly, the link for the YAC512 datasheet is down. But from the OPL3 datasheet you can see there's no data line to trigger a synchronous update of the DAC channels. So maybe this means that the update of the two channels happens with a slight delay between the two. Depending on the serial clock frequency between OPL3 and DAC which isn't specified in the datasheet, I guess that delay might be up to a quarter of the 49kHz sampling frequency. So when sampling this stereo signal with e.g. 44.1kHz, this might lead to some samples being off one cycle. Note that this is only a theory, maybe the DAC's (unavailable) datasheet would mention that the outputs are only updated after the second value update...

Anyway, having a minimal phase difference between the two channels does lead to the result that your phase cancellation method a) doesn't cancel out the sound completely (-30dB is still very much) and b) leads to a so-called "comb filtering" effect, i.e. some frequencies are cancelled out more than others. http://en.wikipedia.org/wiki/Comb_filter. This effect can be imitated by either a comb filter FX plugin (surprise ;)) or, more or less, by a combination of notch filters.

I think it's safe to say that such effects of the OPL3 are definitely not made on purpose but maybe simply to save costs by omitting a signal line for synchronous update. This was a budget mass product. And if you want to have that effect, the signal quality would degrade much less if you use FX instead of your cancelling method.

Oh and btw, CL have my full sympathy for not answering extensive and philosophical emails about minor design flaws of a fifteen year old chipset they didn't even develop themselves :D

S

[stryd_one]

there's no data line to trigger a synchronous update of the DAC channels.

I thought that might be it.... Thanks seppo!

[Green Xenon]

So when sampling this stereo signal with e.g. 44.1kHz, this might lead to some samples being off one cycle.

Nope. The sample-rate doesn't have anything to do with it because I can get the same effect using my analog karaoke voice canceller.

This device is purely analog [hence it does not alias] and doesn't require any power supply to work.

It's frequency response is likely higher than highest-frequency signal the OPL3 can output undistorted.

[Green Xenon]

Try this:

Open the original stereo file

Upsample it to 88200 (twice the rate). Use the best interpolation available in your editor.

Add ONE sample to the beginning of the right channel

flip one channel and mix them together

Compare that with the same thing without the one sample added. Are you getting it yet? ;)

Sorry to be such a pain. I don't understand the 'beginning' of "Add ONE sample to the *beginning* of the right channel" and "Compare that with the same thing *without* the one sample added".

I use Adobe Audition 1.5 to change the sample-rate. I haven't yet upsampled because I not sure whether or not to use the "Pre/Post Filter" option prior to changing the sample-rate.

Once again, sorry if you feel this is annoying.

I have a disability called Asperger's syndrome, you can read about it here:

http://www.midibox.org/forum/index.php/topic,11515.msg91685.html#msg91685

[NV]

Nope. The sample-rate doesn't have anything to do with it because I can get the same effect using my analog karaoke voice canceller.

This device is purely analog [hence it does not alias] and doesn't require any power supply to work.

It's frequency response is likely higher than highest-frequency signal the OPL3 can output undistorted.

Simply because you are running the device out into the analogue domain does not eliminate the effects that digital sample and bit rate may have on the audio. Once the audio leaves the digital realm it carries with it whatever artifacts it may have picked up along the way. For example, if you reduce the bit rate of a sample from a typical 16 bits down to a still dynamic 12 bits, you will notice the sound becomes a bit more "crunchy" or "lo-fi." Lowering the bit rate beyond that will have even more of a dramatic effect. The fact that you can even hear this effect coming out of your speakers is indicative that digital artifacts are still very much audible in analogue domains.

The fact that the karaoke device you mentioned does not alias is beside the point. Aliasing is a phenomenon that is independent from the phase shift we are discussing.

[Green Xenon]

Simply because you are running the device out into the analogue domain does not eliminate the effects that digital sample and bit rate may have on the audio. Once the audio leaves the digital realm it carries with it whatever artifacts it may have picked up along the way. For example, if you reduce the bit rate of a sample from a typical 16 bits down to a still dynamic 12 bits, you will notice the sound becomes a bit more "crunchy" or "lo-fi." Lowering the bit rate beyond that will have even more of a dramatic effect. The fact that you can even hear this effect coming out of your speakers is indicative that digital artifacts are still very much audible in analogue domains.

In no part of my signal chain is the sample-rate or bit-resolution altered.

[seppoman]

In no part of my signal chain is the sample-rate or bit-resolution altered.

oh yes - it starts with the fact that the OPL3 uses an unusual sample rate of 49.something kHz so when you record it back with a different sample rate, the signal isn't the same anymore. Together with the two channel samples possibly coming out of the OPL3 with a slight delay,  a few things happening in the analog domain, then sampling it back with a different rate, maybe even using the input of the SB16(?) which has a really BAD CODEC could among other things lead to additional time offset if e.g. the codec also samples the cannels one after another and maybe uses another channel order for that... So it's quite possible all these digital and analog things happening to the signal would lead to an offset of one sample.

Anyway, I think stryd has proven that this is the case on your recording. If you happen to like what this does to the sound, you should check the stuff I wrote about comb filtering and find a way to reproduce it in a more hifi way than voice cancellation is capable of, rather than using up all your time with questioning the why and how over again :)

S

[Green Xenon]

oh yes - it starts with the fact that the OPL3 uses an unusual sample rate of 49.something kHz so when you record it back with a different sample rate, the signal isn't the same anymore. Together with the two channel samples possibly coming out of the OPL3 with a slight delay,  a few things happening in the analog domain, then sampling it back with a different rate, maybe even using the input of the SB16(?)

If the effect is caused by a different sample-rate, then why does that effect occur also when I play the OPL3 [without recording it] in real-time through the analog voice-canceller?

I could play the MIDI file through the OPL3 synth directly and on-the-fly without recording it to a Wave file. If I play the sound through a voice-canceller [which obviously has no sample-rate because its analog], the effect on the audio is the same as if I use audio software to phase-invert one stereo channel and combine it with the other stereo channel.

BTW, if by "input of the SB16", you mean the "line in" of the SB16 card, then no it cannot be contributing to any of the effects I described because I don't use the "line in" or even "mic in" for that matter. When recording from MIDI to Wave, all processing is done within the PC and sound card itself. Also, when recording directly from OPL3 to a Wave file, the signal remains digital throughout.

[NV]

If the effect is caused by a different sample-rate, then why does that effect occur also when I play the OPL3 [without recording it] in real-time through the analog voice-canceller?

Because the OPL3 is a digital synthesizer. It creates the waveforms using digital technology and then the DAC chip (YAC512) converts it into analogue so that you may hear it.

I could play the MIDI file through the OPL3 synth directly and on-the-fly without recording it to a Wave file. If I play the sound through a voice-canceller [which obviously has no sample-rate because its analog], the effect on the audio is the same as if I use audio software to phase-invert one stereo channel and combine it with the other stereo channel.

Exactly, that was my earlier point - your karaoke device has nothing to do with the sample or bit rate, and therefore has nothing to do with aliasing, which in turn has nothing to do with phasing. Your karaoke device does not affect the sample rate of the original file at all and therefore has nothing to do with the effect described here; if you can hear it on your computer speakers, you'll hear it through the karaoke device just the same. As soon as you can hear it the audio is analogue, no matter where it is coming from. It is impossible to hear digital audio without it first being converted into analogue.

The phase shift inherent to the file is something which exists whether it is in the digital realm OR the analogue realm. The stereo file you provided was a digital file and is therefore segmented into a sample rate. The individual samples of the left channel of the file in question are ever so slightly ahead of the right channel. The sound you are hearing when you invert the phase is a direct effect of this sample differential. The phase shift occurs in the digital realm, yes, but the fact that you can even hear it shows that the digital effects are manifesting themselves in the analogue realm where you can physically listen to it. You cannot hear digital audio since it is only binary code. You can hear it only once it has been converted to analogue audio. However, any digital artifacts that may affect the audio while it is still in the digital stage are still present once it has been converted to an analogue signal.

I think much of your difficulty in grasping this concept is related to a misunderstanding of how digital audio works. I think Stryd has proven his point somewhat definitively. As Seppoman stated earlier, if you like the effect of inverting the phase of one channel then that is perfectly fine, but he and Stryd are both very much correct with their comments regarding the filtering of frequencies in the signal achieving a similar if not identical effect, sans the additional noise. If you do not believe them for whatever reason, then try Stryd's earlier suggested upsampling experiment. All the information you need to do so should be in the Audition manual.