GUIDE: Stereo to 5-Channel Surround

@ kempfand(Andreas),Eye of Horus(EOH cos i don't know your name) and all:
first: i can read but can't understand the last posts(and lots more).
second: in my simple(and newby)opinion,this thread in the one of the most importants threads in the whole world about audio.(if don't the best of all threads about audio).
--->Please, don't "fight"(i can't find better wordds)to turn this thread(i can call magnific thread)in a dangerous place to be "closed" by any mod!
ps: i'm not against anyone,i'm just "afraid" about this thread!
ah...and sorry for bore!
thanks but..please,this thread don't have "price",money can't buy!

[kempfand]

I don't care if comparing one layout to another results in a certain plotting on CATT. I plotted the layouts on an Adobe Illustrator file during my lunch break at work. Unfortunately, I didn't think to send the file home or I'd post more diagrams. The whole point is, it doesn't really matter what the plots look like, what matters is the sound!

I really don't care too much about scientific plots myself, but after having done listening tested for dozens of decoding rigs and see how they plot in CATT, I came to the conclusion that CATT is a phantastic tool to work on Ambisonics (and other things), and that the plots you get there are most often WYSIWYG. In the end, I fully agree, the important thing is how things sound.

We tested the pre-set decoding rigs and custom made decodes from Emigrator, CATT, ambidec.exe, VVMix (exe & VST), signed filters (using impulse responses similar to the ones published by Angelo Farina), and the conclusion was always: Pentagon (even if people have a ITU 5.1 in their living room).

For fairness reasons, I clearly mentioned (a) that I do not exclude that there are better decoding set-ups, (b) that even amongst the experts, there is agreement that this is still 'subject to discussion' (cf. Farina's presentation at recent AES in Berlin), and (c) that people have 'stong feelings' about this, in that they put patent protection to specific Amisonic decoding schemes. For (c), the most prominent example is Gerzon's Vienna decoding scheme, which is available in the high-end "$$$" amps by Meridian (and they do very very well, at least according to the "Verdi" listening tests).

I will explore the new bidules more, but as I said in my initial comments, they seemed disappointing to me. If others find them to be good, then great, but I don't feel I should be jumped on and told to shut up because I'm a newbie in this forum just because I might disagree.

No problem if you found them disappointing. As mentioned, I value your and anyone's feedback, and I think there is room for improvement. Given today means of Forums and commication across geographies and time zones, we can achieve things that were not possible 10 years back. Might sound 'generic', but I truely believe in this.

I may be a newbie here and I may not have a scientific background, but I've been a music fanatic for years and have been playing with various concoctions of homemade surround sound for 30 years.

I think (hope so) I made it clear that your and everyone's feedback is welcome. I have also seen that you have a background with surround methods (otherwise I wouldn't have entered this discussion at all).

I know what sounds good to my ears, and that's all that matters to me. If you produce something that does sound good to me, I'll be the first one to congratulate you, but if there's a problem with it, I'll also say so. As I said in my postings after testing the new bidules, let's look at them to see if improvements can be made. I'm already working on some improvements and extensions to UpMix Studio and I hope you'll give it a critical listen once I post it.

I believe in 'free markets' in this aspects, so let the readers & users of thses methods judge. As indicated, I am contributing here for competitive reasons (if I would, I should make my licing out of this). Off-topic remark: I'm not a believer in 'free markets' when it concerns certain social aspects, but this is a whole topic for itself and the wrong place to discuss (still couldn't resist making that statement).

Now yes, let close this discussion and have fun!

Yes.

To quote member "Shayne": Peace.

Andreas

[kempfand]

@ jorel: Thanks for your remarks. I really value, especailly from someone who has >750 posts and therefore a proven tackrecord

We'll try so.

Kind regards,

Andreas

[trooper11]

@Eye of Horus

hey, yeah youve helped me out alot in the past and i really appreciate it, as far as looking back for info on the thre new bidules, i just kind of got overwhelmed that i would have to read 29 pages to search out info on it, but i guess ill just have to to get links to things, i just wish there was an easier way of finding things is all lol.

you and kemp have done a great job wiht this and i appreciate the work you guys have done, its really been a help for me, sorry if my questions annoyed you guys. keep up the good work.

hey maybe there should be a new sticky devoted just to updates to the method, like guides and bidules, since its so hard to sift them out from this huge thread. you think that would be a good idea?

oh and i just wanted to know what you guys ,EOH and Kemp, use to go to ac3 tracks? or maybe you dont do taht at all. i was thinking of just using besweet to go to ac3, i just wanted to know if im sacrificing quality by using besweet versus another app.

and alos, i read back through the thread and in your guide for hte new ones oyu state that you should set 16bit in the recorder when for the SAd method oyu said 32bit, is there a reason you changed this? also, should i be using 48000 sample when my wav source is 48000?

[ursamtl]

Originally posted by trooper11
hey maybe there should be a new sticky devoted just to updates to the method, like guides and bidules, since its so hard to sift them out from this huge thread. you think that would be a good idea?

Yes, I think this is needed. Daphy and Andy have done a great job putting everything together on the NeedfulThings server, but without some of the context, all the stuff on there can be overwhelming. There is a PDF with the thread discussions, but it still takes some detective work to locate the right information sometimes.

[ursamtl]

Originally posted by kempfand
For fairness reasons, I clearly mentioned (a) that I do not exclude that there are better decoding set-ups, (b) that even amongst the experts, there is agreement that this is still 'subject to discussion' (cf. Farina's presentation at recent AES in Berlin), and (c) that people have 'stong feelings' about this, in that they put patent protection to specific Amisonic decoding schemes. For (c), the most prominent example is Gerzon's Vienna decoding scheme, which is available in the high-end "$$$" amps by Meridian (and they do very very well, at least according to the "Verdi" listening tests).

I think we'll have to agree to disagree on this point, but you see as I mentioned earlier, because I don't see what we're doing as strict, true Ambisonics, it's hard to say that any one rid is the best. Your tests have led you to believe the Pentagon is best. I have tried rotated Pentagons, unrotated (with no center, just a huge, wide surround layout), modified them to widen the Y signals, pumped a stereo wide signal into the input--everything I could think of but both on my PC and my full surround systems, the result sounded small and boxy with some bad phasing happening. The ambiophonics front bidule you posted improved this somewhat (even with my speakers at +-30° instead of 10). then when I took Panorama and hooked it to Emigrator set to Octagon2 with phantom speakers for the three missing speakers, the sound really started to open up. Once I did the bidule groups with the actual formulas instead of the VSTs, it got better. Boosting the Y signal last weekend added even more. With UpMix Studio I am now close to the big, wide commercial sound I heard on true 5.1 mixes but without it sounding gimicky.

I'm listening to some instrumental music right now through your new instrumental bidule, some acoustic guitar. It sounds quite pleasant for this type of music although it's missing definition and accuracy on the individual notes, sort of like smearing them together. I tried it on music with bass and drums and the bottom was really missing. Perhaps crossing over and applying only to the mids might work because the highs get smeared and the lows just aren't there.

A little earlier I was testing the vocal center bidule. I find that I have to crank the Gain Center up to +10 to get a vocal that isn't overwhelmed by it's own reverb. I will say that once I did get the center level high enough, this bidule sounds better than the other two. I know I said the vocal non center one was my favorite initially, but once I got rid of the doubled percusion sounds by backing off the delays, the soundfield spread seemed quite unnatural with cymbals constantly playing behind my right shoulder. It was kind of weird.

My only concern about the vocal center is the same concern I had about my first MatrixMixerEmulator and the Gerzon bidules, that is putting so much dependancy on the fullrange sound of a center speaker. A lot of center speakers are not designed for solid bass for example, they're optimized for dialog. I'm just wondering what your thoughts are on that.

Anyway, que l'on continue...

[KpeX]

@ all

Please keep this discussion as professional and civil as possible. This is an audio processing thread unlike anything else on the web IMO, and I would hate to see it degrade. We have some excellent minds working on these bidules and 5.1 methods, and I ask that you respect each other's intelligence and to respect others' work when posting. As always, thinking before speaking will be greatly appreciated. Cheers,

[Eye of Horus]

Originally posted by trooper11
@Eye of Horus

hey, yeah youve helped me out alot in the past and i really appreciate it, as far as looking back for info on the thre new bidules, i just kind of got overwhelmed that i would have to read 29 pages to search out info on it, but i guess ill just have to to get links to things, i just wish there was an easier way of finding things is all lol.

I agree with you on the search aspect. The search routine used here is not the ideal one :-)

you and kemp have done a great job wiht this and i appreciate the work you guys have done, its really been a help for me, sorry if my questions annoyed you guys. keep up the good work.

Thanks !! We try :-) (that is : if we are not too much distracted by useless discussions :-))))

hey maybe there should be a new sticky devoted just to updates to the method, like guides and bidules, since its so hard to sift them out from this huge thread. you think that would be a good idea?

That's upto the moderators......... who BTW don't seem very active lately ! in the past I remeber that certain psots would have been deleted immedietely :-)

oh and i just wanted to know what you guys ,EOH and Kemp, use to go to ac3 tracks? or maybe you dont do taht at all. i was thinking of just using besweet to go to ac3, i just wanted to know if im sacrificing quality by using besweet versus another app.

Yes, you lose quality ! the freeware AC3 encode doesn't compete with a commercial one like softencode or the one in Scenarist. If you need advice about the settings of commercial products, please contact me privately. I don't think this is the place, but on the other hand : search :-) I know there are some topics about running Softencode in commandline mode and you can find settings there....

and alos, i read back through the thread and in your guide for hte new ones oyu state that you should set 16bit in the recorder when for the SAd method oyu said 32bit, is there a reason you changed this?

Yes...... the reason is that Surcode does internally the same (32 bits to 16 bits). 32 bits WAV's don't play on regular players (like MS Mediaplayer) and they take a lot of space. People with FAT32 file systems won't be able to convert long records in one go in 32 bits, but with 16 bits they can convert longer ones. (Actually 2048/3 = about 682 MB long stereo CD's). We thought this way the whole process would be easier for a lot !

also, should i be using 48000 sample when my wav source is 48000?

Yes ! (And don't forget to adjust Plogue Bidule to 48000 hz too !)

kind regards,

EoH

[Eye of Horus]

Originally posted by KpeX
@ all

Please keep this discussion as professional and civil as possible. This is an audio processing thread unlike anything else on the web IMO, and I would hate to see it degrade. We have some excellent minds working on these bidules and 5.1 methods, and I ask that you respect each other's intelligence and to respect others' work when posting. As always, thinking before speaking will be greatly appreciated. Cheers,

Thanks for the input KpeX !
I fully agree and only wished you'd come up earlier :-)

Anyway : are you still in the conversion "business" ?
Because we never heard any comment from you about our SAD51inBidule, which was based on your conversion method !


kind regards,

EoH

[kempfand]

Originally posted by trooper11
also, i read back through the thread and in your guide for hte new ones oyu state that you should set 16bit in the recorder when for the SAd method oyu said 32bit, is there a reason you changed this? also, should i be using 48000 sample when my wav source is 48000?

and

Originally posted by Eye of Horus
Yes...... the reason is that Surcode does internally the same (32 bits to 16 bits). 32 bits WAV's don't play on regular players (like MS Mediaplayer) and they take a lot of space. People with FAT32 file systems won't be able to convert long records in one go in 32 bits, but with 16 bits they can convert longer ones. (Actually 2048/3 = about 682 MB long stereo CD's). We thought this way the whole process would be easier for a lot !

Another important reason is, that the process for 3 new bidules includes a normalization on the L,R,C,SL,SR speakers. Note that it is not a normalizaion over the 6-channel WAV (output from Bidule), but a normalization on each of the speakers individually.

This is currently not possibly in a "one go" commandline in BeSweet, so you have to first demux, and then normalize each of the channels.


More on the same topic:

Originally posted by ursamtl
I'm listening to some instrumental music right now through your new instrumental bidule, some acoustic guitar. It sounds quite pleasant for this type of music although it's missing definition and accuracy on the individual notes, sort of like smearing them together. I tried it on music with bass and drums and the bottom was really missing. Perhaps crossing over and applying only to the mids might work because the highs get smeared and the lows just aren't there.

A little earlier I was testing the vocal center bidule. I find that I have to crank the Gain Center up to +10 to get a vocal that isn't overwhelmed by it's own reverb. I will say that once I did get the center level high enough, this bidule sounds better than the other two. I know I said the vocal non center one was my favorite initially, but once I got rid of the doubled percusion sounds by backing off the delays, the soundfield spread seemed quite unnatural with cymbals constantly playing behind my right shoulder. It was kind of weird.

I guess you just took the bidules and listened via the ASIO device on the speakers of the PC. That will give bad results. The bidules, as published, are intended to work reasonably well if you follow up with the normalization step, and then listen to the output.

This might sound over-complicated, but we found that you can tweak the gains for the speakers (L+R, C, SL+SR) to work well for one piece of music, whereas it would not work well for the next piece of music.

Normalization seemed a good compromise also for publishing here.

Originally posted by ursamtl
I think we'll have to agree to disagree on this point(of how to best decode), but you see as I mentioned earlier, because I don't see what we're doing as strict, true Ambisonics, it's hard to say that any one rid is the best. Your tests have led you to believe the Pentagon is best. I have tried rotated Pentagons, unrotated (with no center, just a huge, wide surround layout), modified them to widen the Y signals, pumped a stereo wide signal into the input--everything I could think of but both on my PC and my full surround systems, the result sounded small and boxy with some bad phasing happening. The ambiophonics front bidule you posted improved this somewhat (even with my speakers at +-30° instead of 10). then when I took Panorama and hooked it to Emigrator set to Octagon2 with phantom speakers for the three missing speakers, the sound really started to open up. Once I did the bidule groups with the actual formulas instead of the VSTs, it got better.

Emigrator has some shortcomings, which I understand are significantly improved with the commercial version (DecoPro). The standard decodes by Richard Furse, as done by AmbiDec have other shortcomings (as one example, they use no shelf filters). I personally had my best decodes with using signed filters. In practical terms, one can either use CATT, or 2 parallel instances of Voxengo's pristine space since you need 15 concurrent convolutions (3 for each of W,X,Y times 5 speakers).

The newest research (driven by Farina, Malham, and others) even explores using one of the 5 speakers for vertical information (Z-axis). Even if the feeding source is 'just' stereo, information for the 3rd dimension (Z-axis) might be re-created by convolution with the a B-Format pulse. This is still in the research status, but I think we will see a whole range of new decoding options over the next couple of years. Lot's of things to play with and have fun

Regards,
Andreas

[ursamtl]

Originally posted by kempfand
I guess you just took the bidules and listened via the ASIO device on the speakers of the PC. That will give bad results. The bidules, as published, are intended to work reasonably well if you follow up with the normalization step, and then listen to the output.

That's an excellent point and yes I have overlooked it. Before I take the time to burn the CD, I like to tweak the sound as much as possible first. I just found it strange that the center required such a boost in the bidule to begin with.

As I mentioned last night in a message, one concern I have with these new bidules as with the LCR upmix bidules is that the front and rear left and right signals are getting all midrange and high frequencies and one speaker, the center, gets almost all the bass info. Most surround systems don't have a center speaker with as good a bass response as the front left and rights. The center speaker is traditionally designed for dialog. Yes, a high-end system should have a center speaker with good frequency response, but not everyone doing these decodes has a high-end system! If one has a subwoofer and bass management on the amp, this certainly improves things, but it's a waste of the range capabilities of the left and right speakers. Normalization won't change the frequency content of the left and right channels in a bidule, it will only increase or decrease levels as required. Anyway, just a thought.

By the way, try the following for an experiment: take a stereo difference signal and run it through an instance of Stereo Touch. Connect that to the fronts. Do the same with a second instance of Stereo touch and connect it to the rears. Take a stereo sum signal, reduce it by 3dB and connect it to the center. The result should sound familiar. Let's call it "Mr. Hafler's Repurposing Bidule"

The newest research (driven by Farina, Malham, and others) even explores using one of the 5 speakers for vertical information (Z-axis). Even if the feeding source is 'just' stereo, information for the 3rd dimension (Z-axis) might be re-created by convolution with the a B-Format pulse. This is still in the research status, but I think we will see a whole range of new decoding options over the next couple of years. Lot's of things to play with and have fun

Yes that does sound interesting. The only problem I have with this is like with the Ambiophonics +-10° bidule, having to move speakers from their usual positions. I regularly rent or purchase movies on DVD and enjoy 5.1 surround sound immensely. My TV room has a somewhat challenging shape and it took me awhile to get close to a proper ITU-5.1 setup fit in there. I don't want to start moving stuff around! However, if the Z info is implemented in the surrounds, well that could be interesting, since mine are about a meter above ear level.

Tchuss
Steve.

[trooper11]

@EOH

thanks, i sent you a pm btw

[ursamtl]

Just a note for those using SIR, Christian Knufinke released a new version yesterday.

The web site says v 1.007, but the download is actually v1.008.

http://www.knufinke.de/sir/index_en.html


Regards,
Steve.

[ursamtl]

For those experimenting with designing new surround bidules, I'm posting a couple of links to useful articles I've found while surfing. Although these both relate to mixing or programming for surround sound, they contain information that can help and considerations to keep in mind when designing surround sound bidules or plugins.

This first one has some interesting points regarding use of the LFE and center channels, two things that have come up in this thread:

• 5.1-channel Sound: From the studio to your home theater by Leslie Shapiro

While this second article is more programming related and deals with DPL, it's still interesting and provides some discussion of sound cancellation.

• Dolby Pro-Logic Encoding by Robert Basler

Anyone else with useful links to practical surround-related information?

Regards,
Steve.

Hello everybody:

Through a friend of mine, I can heard the post over "Rebuild The Wall".
'Till now, I still surprise with the overall sound. I think it is a forward step to right direction. Here are dinamics with no distortion, well focused any instrumentals, tonnaly very well balanced and all of this produces a typical sense of happiness. God job, guys.

Only I want make a question. IMHO, the conversion lacks of voice definition. It seems to me that the voices are ONE STEP BACKWARD over the other instrumentals. This causes an effect of opacity over all the voices. But, if we are hearding only voices (there are a piece so), these voices sounding really great. If we go to ear the instrumentals (not the voices), that's the same: they are great. But if two of the worlds are joined, this a dissapointment. I can't ear clearly the voices. This effect can make resolve with the new values of the gains? I will say, if the values of center are aumented, the voices are more presents?

Thanks and Good Luck.

[ursamtl]

Originally posted by LAYLA1970
Only I want make a question. IMHO, the conversion lacks of voice definition. It seems to me that the voices are ONE STEP BACKWARD over the other instrumentals. This causes an effect of opacity over all the voices. But, if we are hearding only voices (there are a piece so), these voices sounding really great. If we go to ear the instrumentals (not the voices), that's the same: they are great. But if two of the worlds are joined, this a dissapointment. I can't ear clearly the voices. This effect can make resolve with the new values of the gains? I will say, if the values of center are aumented, the voices are more presents?

Hi Layla,

I had the same problem. Kempfand pointed out that I wasn't doing the normalize step. Did you try that? This did help my results a bit. I also modified the center feed to the standard -3dB L+R signal. To accomplish this, Ctrl+click on the "Preservation_3_x_2" group in the center and change the multiplier from 0.4511 to 0.7071. This number produces a -3dB gain, whereas 0.4511 is close to -7dB.

While you are in the group, you can also remove the the unused left and right multiplier, constant, and connection objects, since they are not hooked up to anything but still seem to use CPU resources. After I got rid of them, my CPU usage dropped by an average of 2-3%. I know that is not a big drop, but every little bit helps!

Regards,
Steve.

[trooper11]

@EOH i sent you a couple pm's



right now im experimenting with the 3 new bidules and the SAD method. right now though, the SAD mehtod is giving better results, but i think thats my fualt. using hte instrumental bidule, the result was lower in sound then in the SAD method. I think it may be becuase before i loaded the file into the bidule, i had already normalized it, converting it to 32bit as well. As i followed the guide, i know it was normalized again by besweet, so could this cuase the sound to seem lower, dropping out?

ok and my other question for you guys, which of these bidules do you think would work well with audio tracks? By audio i mean movie audio that i want to take to 5.1. Or maybe none of these are suited for that.

[Shayne]

Originally posted by kempfand
and

Another important reason is, that the process for 3 new bidules includes a normalization on the L,R,C,SL,SR speakers. Note that it is not a normalizaion over the 6-channel WAV (output from Bidule), but a normalization on each of the speakers individually.

Regards,
Andreas

normalizing left and right individually would loose the initial stereo would it not? I feel the l and r whether it be front or rear should be grouped. Its the F R C LFE groups that would be nice to be able to normalize independant on the fly but then i would be Happy to just isolate the lfe.

Peace

[ursamtl]

Originally posted by Shayne
normalizing left and right individually would loose the initial stereo would it not? I feel the l and r whether it be front or rear should be grouped. Its the F R C LFE groups that would be nice to be able to normalize independant on the fly but then i would be Happy to just isolate the lfe.

Peace

To my understanding, all normalizing does is scan a file for the highest peak level and then either boosts or attenuates the entire file in a linear fashion so that this highest peak reaches a target level.

In the case of the 3 bidules, the "instrumental" bidule has the same signal going to both fronts, and the same signal going to both backs. To check this, set up a binary operator in the bidule as a minus and then hook both signals up to the operator's input and connect the output to one channel of your sound card. If nothing comes out, the signals are the same. Therefore, normalizing these channels independantly will give balanced results for the front between the left and right and the same for the rear.

For the "vocal-center" bidule, the same signal (L-R) is sent to Voxengo to generate a pseudo stereo signal. (Yes, one is a R-L, but if you invert it, you get the same as L-R. You can check this the same way as above). In this bidule, the front left and rear right channels get the same signal. The front right and rear left get the same signal. Since these signals are all generated from duplicate Stereo Touch instances, the levels are basically the same and thus normalizing all these channels independantly should keep things balanced.

The "vocal-non center and instrumental" bidule is a bit more sophisticated in that the improvement groups contain filters as well as sum and difference combinations to create four different channels, but these are all at about the same level. If you try the channel subtracting test I mentioned above on all combinations of these channels, you'll always still hear something (indicating that there is some difference), but the difference levels are low and consistent, so normalizing these individually shouldn't be an issue. I haven't dug into this bidule any further to analyse what's going on because I don't like what it does to music with percussive sounds and I'm busy developing something else I find more interesting.

In the end, follow the bidule author's advice. Andreas said that he got the best results by normalizing the channels individually, so try it like that. Then if you don't like the results, try something else. Half the fun is in the trying!

Have a good weekend!
Steve.

[Shayne]

Originally posted by ursamtl
To my understanding, all normalizing does is scan a file for the highest peak level and then either boosts or attenuates the entire file in a linear fashion so that this highest peak reaches a target level.

In the case of the 3 bidules, the "instrumental" bidule has the same signal going to both fronts, and the same signal going to both backs.

Steve.

Your understanding of normalization would be correct. It finds the highs peak and increases all data points by the same amount(dbs).

If the 3 new bidules merge the original left and right and loose all stereo i would have to stand by my post above and wonder if this is right?

Peace