L - R (difference) Processing?

[GeneSavage]

When AM stereo came out, I remember big discussions in audio processing circles on the need of a processor that handled the signal in sum and difference mode instead of standard stereo left and right mode because of loudness issues on mono radios.

Several questions:

Did this mean that if you had something in only one channel, it would come across as loud as something in both channels?

What are the disadvantages to processing sum / difference?

Is such an audio processor available for computer audio?

I ask because I run an internet station, & today I am at a location where, for very boring reasons, I am listening in mono.

A recording just played that had voices exclusively in the left and exclusively in the right channels at different times.

The volume certainly sounded to be about halved!

I'm currently using a software-based audio processor called "Sound Solution," a slow AGC followed by 5 bands of compression, 5 bands of expansion, 5 bands of limiting, a dual band final loudness processor, and a peak clipper:

http://www.winamp.com/plugins/details/120741

To my knowledge, it does not chain the stereo channels together; I was listening through headphones yesterday to an acoustic piece with lots of dynamic range that had the lead singer swirling around the stage (as the two channels independently bounced up and down in volume)!

I realize 99.99% of my audience will NEVER listen to an internet radio station in mono, but for that .01%, I'd like to know if there's a way to correct the volume issues with discrete channel content without doing damage to the stereo signal... especially if it's a software solution.

I figured if anybody would know the answers to my questions, it would be this group!!! 8)

Thanks for your help.

 

[RadioMan203]

For the 7+ years I ran CapitalRadio.us there was about three times I monitored the station in mono. I don't think you really have much to worry about.

As far as a software solution, I think you'd be hard pressed to find one. There are hardware boxes such as the CRL AM Stereo processors, the Ariane (which I would have loved to get for the high school FM I'm building except it's out of their budget) and perhaps a couple of others. Another processor that does L-R processing that works on a totally different platform is the ModSci Stereomaxx, which makes your stereo platform wider. It's transparent in mono as it's augmented signal is suppose be phase cancelled.

Other than hardware I don't know of any software emulated processor that specialized in l-r processing.

 

[whit979]

Is such an audio processor available for computer audio?

The Vorsis VP-8 uses sum/diff and it has an MP3 mode on it.

 

[gHz]

The Sequel....ie Digital Ariane is essentially a software L-R processor. (L+R too) Of course David Reaves et al, own the software algorythms that produce the stunning results. I do not think you will find their DSP code on a shareware site.

I am not aware of any credible L-R software processors. That does not mean that none exist.

If you could break the audio out analog for a piece, an old Stereo Maxx might be the answer to your search. Before the Sequel was available, I did exactly that on a totally digital broadcast chain to take advantage of the Ariannes L-R processing. The chain went D>A, Arianne, A>D right after the EAS digital converter insert point. It did nothing to harm the digital signal, and it added significant stereo performance enhancement that did not adversely affect the mono sum at all. There was significant expense in doing it that way.

I will agree with you....every stereo audio image should maintain integrity in mono. As soon as you begin to think otherwise, you will be bit in the butt. Trust me.....I made that mistake....once.

Gary Z CE
Cleveland

 

[Sgeirk]

Stereomaxx's are very good for the internet. If that's one fewer on an FM station, all the better.

 

[cgould]

Hi there

Stereomaxx would actually be pretty lousy for internet broadcasting as it loads up the L-R region with phase and time delayed audio. This will cause the CODEC to have to work really hard on the L-R information, robbing bits away from L+R, causing excessive artifacts.

The reason stereo CODECs at typical internet bit rates use "joint stereo" is because there is usually far less L-R information than L+R in most stereo programs, and the CODEC can use this as an advantage to place the bulk of the available bits in the mono region to improve perceived audio quality.

For the same reasons stated in the first paragraph, the thing was pretty lousy for FM Stereo too. Gave you bizarre sounding stereo (IMO), and lots of multipath to go with it! Some people still really love that thing, though.

-Cornelius

 

[rrsounds]

The Sequel....ie Digital Ariane is essentially a software L-R processor. (L+R too) Of course David Reaves et al, own the software algorythms that produce the stunning results. I do not think you will find their DSP code on a shareware site.

I am not aware of any credible L-R software processors. That does not mean that none exist.

If you could break the audio out analog for a piece, an old Stereo Maxx might be the answer to your search. Before the Sequel was available, I did exactly that on a totally digital broadcast chain to take advantage of the Ariannes L-R processing. The chain went D>A, Arianne, A>D right after the EAS digital converter insert point. It did nothing to harm the digital signal, and it added significant stereo performance enhancement that did not adversely affect the mono sum at all. There was significant expense in doing it that way.

I will agree with you....every stereo audio image should maintain integrity in mono. As soon as you begin to think otherwise, you will be bit in the butt. Trust me.....I made that mistake....once.

Gary Z CE
Cleveland

There are a few reasons the Ariane typically doesn't get you "in trouble" in mono. I think one of the most important is that you can't drive the L-R input harder than the L+R. In Matrix mode, the Ariane L+R and L-R inputs and thresholds are identical and not adjustable.

Generally as a rule, as long as the L+R stays higher than the L-R you are _relatively_ safe. With conservative processing, i.e., basically civilized compression ratios and system timings, keeping identical thresholds and drive levels tends to _keep_ you out of trouble. One can always speed up the Sequel's L-R timing w/regards to that of the L+R, but you're still safer than with two fully independent compressors.

Personally, I don't like or approve of the Stereomaxx effect, but I admit to being biased in this regard.

Kind Regards,
David

 

[RadioMan203]

Hi there

Stereomaxx would actually be pretty lousy for internet broadcasting as it loads up the L-R region with phase and time delayed audio. This will cause the CODEC to have to work really hard on the L-R information, robbing bits away from L+R, causing excessive artifacts.

The reason stereo CODECs at typical internet bit rates use "joint stereo" is because there is usually far less L-R information than L+R in most stereo programs, and the CODEC can use this as an advantage to place the bulk of the available bits in the mono region to improve perceived audio quality.

For the same reasons stated in the first paragraph, the thing was pretty lousy for FM Stereo too. Gave you bizarre sounding stereo (IMO), and lots of multipath to go with it! Some people still really love that thing, though.

I have to say I actually used a Stereomaxx as part of my webcast for the seven years I had it online and when properly set up it added a wider stereo platform to the station and made it stand out when compared to those running similar programming. I did have to make major adjustments to the separate encoding PC and software to improve the output quality for the transmission bitrate.

I do have to say that my experience with operating Stereomaxx on terrestrial FM can be problematic for those who get a little carried away with the controls as I've noticed excessive multipath induced by its use. Used very gently it can give a pleasing effect. I do believe when people talk about matrix processing these days they are referencing a processor like the Ariane or similar acting unit and not the Stereomaxx.

I do agree with David in that keeping levels in check is paramount and would be easier to perform with a single processor. I'm currently experimenting with my own design for a custom matrix processor for a low-budget facility since they pay me regardless of the project I'm on. It's been a little hit and miss but the system is performing fairly well on the bench.

 

[cgould]

I do have to say that my experience with operating Stereomaxx on terrestrial FM can be problematic for those who get a little carried away with the controls as I've noticed excessive multipath induced by its use. Used very gently it can give a pleasing effect. I do believe when people talk about matrix processing these days they are referencing a processor like the Ariane or similar acting unit and not the Stereomaxx.

I've found them to be problematic and not too nice at any level of adjustment, but hey, that's how my tastes go!

I really tried to like the thing for a better part of a 1.5 years, and found after all the fiddling, I let out a resounding "aahhhh" the day I bypassed it, and just took them out of the chains that had them after that ;-).

If it does what you want, then I guess that's what matters the most!

-Cornelius

 

[rorban]

(1) Did this mean that if you had something in only one channel, it would come across as loud as something in both channels?

(2) What are the disadvantages to processing sum / difference?

(3) Is such an audio processor available for computer audio?

(1) Absent clamping (see my answer to [3] below), yes.

(2) I don't think it's as natural sounding as L/R processing for most program material. This is because of the "single channel sounds as loud as both channels" issue.

(3) Speaking for Orban, all of our digital processors since the 8400 have allowed you to operate the AGC in sum and difference mode. The AGC provides user controls that allow the the maximum amount of sum/difference gain offset to be clamped independently in the "L-R dominates" and "L+R dominates" directions so that you don't get too much pumping of the stereo soundstage width. The main purpose of introducing sum/difference processing was to allow the AGC to automatically reduce the separation of certain oldies (like Beatles records) where the "stereo" is actually multitrack elements instead of a real stereo mix.

In these products, the multiband compressor/limiter operates in L/R mode only. We believe this is preferable because experience has shown that it provides the best loudness and textural consistency from source to source and within sources. In our opinion, running the AGC in sum/difference mode is sufficient to achieve the user's goals and most of the time it is wise to clamp the sum/difference gain offset fairly tightly.

Our main Internet-targeted product is Optimod-PC, whose AGC can be operated sum-and-difference.

 

[GeneSavage]

StereoMaxx.. wow! I have a cassette demo of theirs somewhere that gives impressive, fascinating... occasionally disturbing... results. I'm 99% sure a local CC AOR is still using a StereoMaxx, and every so often I hear something jump out of FAR left or FAR right that just seems a little "too" localized... there was a time I thought I would want one on the air, but hearing it in use, my mind has been changed.

I'm not really looking for stereo "enhancement" per-se; in fact, the processor I currently use has a stereo enhancer I've played with a bit but have never really liked, either. Original separation seems to cause the least "problems" down the chain.

Bob: thank you so much for the direct answers to my questions! Those answers do help, although (of course) they bring up more questions:

1 - In order for something fed to one channel only to sound as loud as something fed to both channels, you would have to drive the L - R harder than the L + R, yes? (150% maybe? Yikes!)

2 - It sounds like you could run an AGE in sum / difference mode and feed that output to a L / R mutil-band processor... that, in fact, that's what the Orban boxes do, yes?

3 - I wonder if anybody knows of any comparisons online (mp3 files?) of how L / R vs. L+R / L-R processing sounds?

I started this thread to solve a problem, but now the mechanics of what we're discussing fascinate me!

(BTW, for the person who replied that mono concerns on a stereo stream are almost irrelevant: I agree... ALMOST. However, I'm the guy who found a computer at work that clicked, crackled and popped on any sample rate besides 44.1 KHz. [I was using mp3Pro and on players without the codec seeing 22.05 KHz.] Was the soundcard or its driver bad? Likely. But the listener doesn't care WHY they can't hear you [or why you don't sound as good] they just know they can hear somebody else clearly & not you.

So knowing someone MIGHT be listening in mono, that it COULD happen, makes it a concern. It was also through the mono box that I discovered a copy of a song that sounded perfectly fine in stereo actually has serious phase issues! So, while it's not going to be the only thing I focus on, it IS a concern for me.)

 

[jesseg]

It's not too uncommon in the music recording industry, to see engineers/producers who are panning stuff dead center, or all left & all right, with the exception of very little stereo reverb/room fx if any (and no early reflections). You might also think that approach can leave a mix sounding like 3 positions, but it's the interactions of everything that really can make a mix sound as wide as any with crazy panning and horrible amounts of effects.

I've even heard of a few studios that have the boards wired so they can only pan hard or center. 8)

My point is, which do you think sounds better on air (regardless of L-R/L+R), translates the best to the average playback system, and sounds great in mono too. A LOT of major albums were mixed this way.

As far as how L-R processing can effect audio that has bad inter-channel phase issues... it partly depends on how the L-R processing is handled. If it's just a M/S matrix, it has to make the problem worse by design if it's adding more L-R. I'm not discounting that there are ways to limit the severity of it, a few have already been mentioned.

 

[GeneSavage]

To clarify, my original reason for asking about L - R was to overcome the limitations of a recording(s) which has dry voices taking turns talking in the left-only or right-only channels.

Listening in mono, these voices were at (about) half the volume of a centered mono source (equal in both channels)...

...the listener listening in mono would hear that the voices on the 1st recording I described were quiet, but would have no idea why.

One friend suggested re-cutting the recording & summing the channels! Well, yes, that fixes the mono listener problem, but now the stereo listeners have no separation...

So, to get it to full volume in mono, would L - R have to be cranked up to 150%? (Ick!)

 

[fm-engineer]

The neat trick with Ariane is you can actually have less L-R than the original recording but still sound wider. There is a spatial effect because the L-R is processed independently. I have witnessed significant improvement with multipath prone stations in western Maryland and PA.

 

[cgould]

To clarify, my original reason for asking about L - R was to overcome the limitations of a recording(s) which has dry voices taking turns talking in the left-only or right-only channels.

Listening in mono, these voices were at (about) half the volume of a centered mono source (equal in both channels)...

...the listener listening in mono would hear that the voices on the 1st recording I described were quiet, but would have no idea why.

One friend suggested re-cutting the recording & summing the channels! Well, yes, that fixes the mono listener problem, but now the stereo listeners have no separation...

So, to get it to full volume in mono, would L - R have to be cranked up to 150%? (Ick!)

Hi, Nightaire!

The reason for the volume drop is the simple nature of stereo. Think of it this way: The left and right channel each make up 50% of the total volume of a mono signal. Together, you get 100%. Either one alone will drop the mono signal by half...which you have already observed.

Messing with the difference channel will do nothing for mono listeners as the L-R processing effect is canceled out totally in mono. What you are looking for is to have something "ride" the mono signal and leaving the difference channel alone. Doing this in a way that solves the mono issue will probably cut the stereo separation for stereo listeners by about 50%, though.

About the best solution I can think of that solves both issues is to just set up a mono stream for the mono folks (which may be what you have - you didn't specify), and dedicate a "mono processor" for that channel. Do not feed the processor dedicated to the mono channel with stereo material. Feed it mono, and viola! You'll find that it will even out these variations from the stereo source material....

-Cornelius

 

[rrsounds]

The neat trick with Ariane is you can actually have less L-R than the original recording but still sound wider. There is a spatial effect because the L-R is processed independently. I have witnessed significant improvement with multipath prone stations in western Maryland and PA.

Exactly. The less you load up the L-R subcarrier with energy, the less likely that you'll get the traditional multipath interference.
By processing the RMS value of the L-R to make it more consistent, then reducing its level, you can achieve the effect fm-engineer describes.

Another FM benefit of the Ariane's multiband matrix mode is that you can reduce the low frequencies in the L-R, which has a useful effect:
Since much of the distortion you hear with multipath interference is inter-modulation, reducing the extreme bass tends to reduce the audibility of the 'gargly' IM that may otherwise occur. This practice also reduces the amount of clipping of the L-R if you are using a composite clipper, further reducing the likelihood of IM from multipath.

You might note that while bass in music is usually mono already, just a dB or two of accidental (or intentional) channel imbalance puts a noticeable amount of bass into the subcarrier. This reduction of low frequency L-R energy by the Ariane can help fix that, if desired.


As for internet processing, the Ariane's L+R bands can never be controlled by the L-R signals while, OTOH, the L-R bands can be slaved to a user-defined limit in their gain relative to the L+R. This generally leads to very good mono-compatibility, as the mono can not be compromised by whatever you do with the L-R.

One caveat: If any L/R linked processing follow the Ariane, you will clearly hear when you have too much L-R: center vocals seem pushed back. But this only results from pretty extreme settings, and it's a simple matter to reduce L-R injection level to the point this effect no longer occurs.


Kind Regards,
David

 

[GeneSavage]

Cornelius -

What you said makes complete sense! No, I have a stereo stream so most people wouldn't hear this "problem" anyway.

You're right, I assume compressing sum & difference would result in an almost "fixed width" effect on all songs, so that narrower ones would be wider, and wider ones would seem to be more narrow... or maybe the difference would be how strong the sum channel was... I'm still learning.

I also got to wondering this morning if the file could be "pre-processed in something like Adobe Audition, but I assume my L + R audio processor would undo anything I "fixed."

As a matter of fact, we're looking as some cheaper hosting that might allow us the cash to set up a mono (dial-up) stream... I'm not sure how we'll process it independently, but you bring up a good point!

 

[cgould]

I'm not sure how we'll process it independently, but you bring up a good point!

This could be done easily by attaching a "summing amplifier" to your main program feed, which will give a mono version of your stereo program audio to feed into the dedicated processor. It can be a stereo audio processor fed with Mono audio from this summing amplifier. You'll just feed the same mono into both left and right inputs.

What you will end up with is a parallel processing setup. One doing stereo as normal, and the other being fed from a tap off of the stereo feed into the mono summing amplifier. You can use one of those RDL "stick on thingies" to take the stereo program channels, and sum them to mono for processor #2.

-Cornelius

 

[RolfTaylor]

Actually, this brings up another related factor. If you plan on a very low bit rate stream it is wise to make it mono in any case.

So if you add a new stream make that a low rate. Mono listeners with sufficient bandwidth will have a choice as to which stream they listen to, and low bandwidth listeners will get the best quality possible at a low bit rate.

Cornelius -

What you said makes complete sense! No, I have a stereo stream so most people wouldn't hear this "problem" anyway.
<SNIP>

As a matter of fact, we're looking as some cheaper hosting that might allow us the cash to set up a mono (dial-up) stream... I'm not sure how we'll process it independently, but you bring up a good point!

 

[GeneSavage]

Rolf,

I agree completely; I don't do stereo below about 64k... actually, our current stereo stream is 96kbps. I was thinking about 24kbps mono for the dial-up stream... I hear some people say dial-up can / should be able to listen to 32kbps (or 40k, or 48k!) but when I was on dial-up that was never my experience; anything over 24k choked.

In regards to an outboard processor, I think it would require a second encoding computer for my current setup. Right now we use one computer to play (a great little program called Station Playlist Studio) and encode (using SpacialAudio's mp3 / mp3Pro encoder). The encoder will let you create as many streams from that one source as your CPU and internet connection will allow.

I don't know that you can process the audio independently, even with a loop and / or a second soundcard, though... I'll have to check with them and see what could be done in that respect.

If I could feed the encoders independent paths using the same source, it would be an easy thing to sum the source & feed an AGC.

...Man, I miss analog...