SPECWEB (Now 2.2)

[Eye of Horus]

Thanks for your explanation.
It makes some things a lot clearer.
In cases like this I always compare the end result with the output from the DTS Neural VST.
Why ? Because it has no settings other than gains. So during the conversion you get immediately an idea how much is steered to each channel and too much center based songs will easily be recognized.
That is : with my default gain settings.
And after your comment about too much center based, I used CenterCutGui and measured the C and S : there is only 1 Db difference between C and S !
A too much center based song would show far lower values in S...... right ?
I think it has something to do with the different versions and mastering of that album. Nice project to try out with 5 different versions ;-)

Question : have you tried out the Nugen Halo Upmix VST ? Separation results are very good after a long (very long) trial of all the different presets and making my own adjustments.
But I really don't understand why it has such a high pricetag and still delivers in all trials I did, artefacts in the rears..............
Compared to the artefacts in SPEC and QSXTPRO, they are not of the swooshing kind, more an aprupt cut off of some tones.
The swooshing kinds are almost gone after processing the mch through WAMINU, but that doesn't work quite well with the output from HALO .

I also want to show my respect here to you, for all the time and effort you put into SPECWEB !

I wonder if it's posiible to build in an option to pause before the limiter and let you put in a treshold level based on the output of first steps before the limiter starts ?
(I know my usage of the Nugen ISL is quite agressive and that's the reason why it's difficult and nearly impossible to do too loud mastered CD's (-10 Db RMS > YUCK !) in WAMINU.)
I also think I have a solution for the "spike" problem, but cannot write it myself nor found something like a VST that can do it for me........
A simple routine I think for you to write ;-)
I will contact you by PM later this week to describe what I think could be a simple solution.....

grtz,

EoH

 

[zeerround]

And after your comment about too much center based, I used CenterCutGui and measured the C and S : there is only 1 Db difference between C and S !
A too much center based song would show far lower values in S...... right ?

Center cut only makes one cut on each side. I think to see the levels low on the left and right edges you would need to use center cut again, with C and the left side as inputs, for instance (this is how slice works), and look at the left side output.

Question : have you tried out the Nugen Halo Upmix VST ? Separation results are very good after a long (very long) trial of all the different presets and making my own adjustments.

No, that's one I haven't tried. I also unfortunately lost most/all of my tools in a hardrive crash. I thought it was a windoze problem and kept trying to recover from that, instead of backing up the drive. Then the drive stopped responding all together (live and learn, I guess). People want > $500 US to try to recover the data... Errrr. A Major impact on my productivity in this hobby (as I can't afford the $500).

swooshing

Swooshing is a known artifact of phasevocoders, the underlying tech in Plogue, Spec, and SpecWeb that convert to/from the time domain to/from the frequency domain, where all the upmix algorithms work. I'm always looking for other solutions, wavlets, HFT, etc. and currently trying to port some code from a center cut clone. Not sure if I can pull that off on my own (It's written in C# and SpecWeb is in C++) but I'll keep trying.

There are also some formulas for a different panning scheme below and above 700 Hz that I want to test to see if it has any effect on swoosh. The paper on that was done in Korea about ten years back, and seems to have a bunch of math errors in it, but via this forum I got some help on correcting those formulas so we'll see. Lots of things to code and test.

In the meantime upsampling before conversion, and judicious use of blends and adjspkr are the tools to combat swoosh.

I also want to show my respect here to you, for all the time and effort you put into SPECWEB !

Thanks!

I wonder if it's possible to build in an option to pause before the limiter and let you put in a threshold level based on the output of first steps before the limiter starts ?

Almost anything is possible, but the way it works now is the threshold you enter in the ini or on the command line is subtracted from the peak output of Zag pass 2 (just before the limiter, ZAG pass one happens in parallel to the first, actual upmix, pass). So, for instance, if you are using match level and your song has it's loudest peak at -10 dB, and you enter -3dB for threshold, the actual threshold sent to the limiter will be -13. Using match level on a quiet song, you can see that in the limiter output. If you don't use matchlevel, then the the peak output of the stage before the limiter will equal your max output zag setting (likely 0 or -0.1 or -0.2 dB) so entering -3 dB threshold will be either -3 or -3.1 or -3.2 dB, respectively.

Hanz mentioned you guys set the limiter in WAMINU using RMS values? What would you like to see?

By the way, I expected you to ask for the ability to use Nugen ISL (or any random VST) and that is also doable (if it works in mrswatson). The parameter values just need to be mapped into 0 to 1 values.

I also think I have a solution for the "spike" problem, but cannot write it myself nor found something like a VST that can do it for me........

I see the limiter as taking care of that, but I would like to see examples of stereo files that produce "spikes" with SpecWeb. Hanz sent me some already converted tracks, but in all cases the "spikes" were clearly at the same frequency as the lower levels around them, so they were just loud parts of the music, not transient spikes, so not an artifact of SpecWeb.

Cheers,
Z

 

[Eye of Horus]

Center cut only makes one cut on each side. I think to see the levels low on the left and right edges you would need to use center cut again, with C and the left side as inputs, for instance (this is how slice works), and look at the left side output.

A typical center based song will show extreme lower values in S....... It was one of the reasons to get rid of Centercut in the WAMINU proces. Now we have different versions with different separators for Center and Stereo Sides.
As separator PT-Tech is used (Unfortunately works only on 44.1 and 48 Khz), or Waves Stereo Center or the 6 channel version which takes outut from methods that contain some artefacts.
Those two separators are fully adjustable, which is a big advantage over CenterCutGui. When however the others don't give a good separation, CenterCutGui is tried too, but it's no longer first choice.

No, that's one I haven't tried. I also unfortunately lost most/all of my tools in a hardrive crash. I thought it was a windoze problem and kept trying to recover from that, instead of backing up the drive. Then the drive stopped responding all together (live and learn, I guess). People want > $500 US to try to recover the data... Errrr. A Major impact on my productivity in this hobby (as I can't afford the $500).

Geezzz, that sucks ! If I can help you getting back some of the major software, just let me know.

Hanz mentioned you guys set the limiter in WAMINU using RMS values? What would you like to see?

Yes, that's almost correct.....
The RMS values are used to get even values in fronts, +3 in C and -3 in LFE and -1 in Rears.
First after the first run, gains are added to get the center channel between -1 and -2 PEAK.
The corresponding RMS level is used to gain the other channels to the above values.
After that the NUGEN ISL 5.1 is used with a peak value of -0.3 and the value for gain is set to the lowest PEAK with 0.4 to 0.6 extra gain.....

Example :

fronts -20 /4 center -17/-1.5 lfe -23/-7 rears -21/-3
This will get a gain of 4.6 Db.... all channels will be at -0.3 PEAK (except in this example the lfe) and RMS will be according to the wanted values above.

By the way, I expected you to ask for the ability to use Nugen ISL (or any random VST) and that is also doable (if it works in mrswatson). The parameter values just need to be mapped into 0 to 1 values.

LOL ! It seems obvious but isn't in my workflow....... I use the SPECWEB output as input for WAMINU to get rid of 90 % of the artefacts and NUGEN is used in THAT proces as last instance......
That is : as long as there are artefacts in the SPECWEB output. when the swooshing problems are gone, then it might be useful to use NUGEN ISL in the SPECWEB workflow. But now it will only make the volume of the swooshings higher and ..... hearable.
I have done so much testing to hear the influence of NUGEN on the voice and could easily gain upto 8 Db on a voice that was already peaking at 0 Db , without hearing any distortion. It's used by quite some famous bands in their productions !

I see the limiter as taking care of that, but I would like to see examples of stereo files that produce "spikes" with SpecWeb. Hanz sent me some already converted tracks, but in all cases the "spikes" were clearly at the same frequency as the lower levels around them, so they were just loud parts of the music, not transient spikes, so not an artifact of SpecWeb.

I have seen those files too. But when the whole conversion is based on peakvalues, those 4 spikes in one of the rearchannels, influence the outcome of all the channels.
Those spikes are 0 Db PEAK, while the channel without spikes is -5 PEAK and both on the same RMS. Not necessary an artefact of SPECWEB ! Every center/sides extractor sometimes gives spikes !
A limiter will IMHO not be the solution .... but I will PM you about a solution I think that can easily be implemented......

grtz,

EoH

 

[zeerround]

The RMS values are used to get even values in fronts, +3 in C and -3 in LFE and -1 in Rears.
First after the first run, gains are added to get the center channel between -1 and -2 PEAK.
The corresponding RMS level is used to gain the other channels to the above values.
After that the NUGEN ISL 5.1 is used with a peak value of -0.3 and the value for gain is set to the lowest PEAK with 0.4 to 0.6 extra gain.....

Example :

fronts -20 /4 center -17/-1.5 lfe -23/-7 rears -21/-3
This will get a gain of 4.6 Db.... all channels will be at -0.3 PEAK (except in this example the lfe) and RMS will be according to the wanted values above.

Sorry but I'm not understanding how any of that applies to ways you would like to set the limiter threshold in SpecWeb, because the word "threshold" isn't anywhere in there, just "gain".

 

[Eye of Horus]

Sorry but I'm not understanding how any of that applies to ways you would like to set the limiter threshold in SpecWeb, because the word "threshold" isn't anywhere in there, just "gain".

That's correct , there is no threshold setting in the NUGEN ISL ! it's nothing more than an explanation how I use it in WAMINU, not saying this is what I want for SPECWEB ;-)
I guess I was more focused on the first line (not quoted !) : "Yes, that's almost correct....." and oversaw the "What would you like to see" ......
But again, I will PM you my idea about that.

 

[seilerbird]

Could Specweb take a mono input source and produce a true stereo version?

 

[zeerround]

Sorry, no.

All the algorithms depend on having a stereo source.

There are techniques for mono to stereo, but to my knowledge they are all manual and time consuming, involving spectral editors where you have to identify each instrument you want to separate out, note by note. Still, people do it.

If you want to know more I can probably get some forum links for you.

 

[seilerbird]

Sorry, no.

All the algorithms depend on having a stereo source.

There are techniques for mono to stereo, but to my knowledge they are all manual and time consuming, involving spectral editors where you have to identify each instrument you want to separate out, note by note. Still, people do it.

If you want to know more I can probably get some forum links for you.

Thanks. I am well aware of spectral editors I was just hoping this would work on a mono source. Thanks.

 

[zeerround]

SpecWeb 1.31 has been released.

Details and download here:

http://www.surroundbyus.com/sbu/viewtopic.php?f=8&t=913

 

[zcftr29]

SpecWeb 1.31 has been released.

Details and download here:

http://www.surroundbyus.com/sbu/viewtopic.php?f=8&t=913

Thanks for your continued hard work...

I think I've come across another bug - I've not tried out 1.31 yet, so it might be fixed... but I think I know what the problem is.

I 've just re-done a conversion of 'The War Of The Worlds - The New Generation' (seperates out very well to Quad by the way) and the longer tracks (12 minutes plus) fail during the 'post-zag-vst' phase of the conversion. If I turn off the usevst switch, the conversions complete ok. I am converting upsampled files from a CD source at 176.4kHz - so the temporary files get pretty big. My guess is that mrswatson has a file size limit of 4Gb due to it being a 32 bit programme. One solution might be to downsample the mch file BEFORE running through zag/mrswatson - unless there is a 64 bit version of mrswatson available

 

[zeerround]

Yeah Mrs. Watson is currently not supporting any file types that support > 4 gig files. I will ping the author again.

Also, I'm trying to port the mastering limiter into spec native, so Mrs Watson won't be needed.

Stay tuned.

 

[zcftr29]

Yeah Mrs. Watson is currently not supporting any file types that support > 4 gig files. I will ping the author again.

Also, I'm trying to port the mastering limiter into spec native, so Mrs Watson won't be needed.

Stay tuned.

SoX may be a viable alternative as it can work with w64 and flac. The 'compand' effect may give the desired results...

 

[zeerround]

Well I'm certainly open to that. The search for a free, good sounding, limiter that supports 6 channels has been a long one. I'd be interested in people's input as to default parameters for sox compand. I'm traveling at the moment and can't really do any critical listening (or any listening in surround).

Generic post processing via parameters sent to mrs watson or sox could be added, such that you could use your own VST(s) and settings or Sox settings.

Having said that, my goals for compression/limiting/remastering built into SpecWeb are two:

1) The overall loudness of the track would be similar to the original stereo (the EU standard loudness measurement, in Sound Forge and others, being the standard). Note that without compression/limiting/remastering the output of Spec/SpecWeb is not "as loud" as the original stereo, even when correctly downmixed to stereo. Also note that it is easy to go overboard with many post processing plugins.

2) Loud but brief (some folks are using the term "spikes") audio samples, particularly in the rears, don't fool ZAG into lowering the channel more than is musically pleasing. This may come "for free" is you accomplish goal number 1. The volume level between the various surround channels should not only be "musically pleasing" but also be true to the original stereo mix. In other words nothing should stand out in the mix, or be buried more in the mix, than in the original stereo (except for the "wow effect" of having things come from their own speaker in surround). Yeah center vocals should pop, as should rear guitars etc. but their actual relative volumes should be the same as in the original stereo mix.

Oh, and I guess there is a number 3) too.

It has to be free/opensource.

 

[himey]

Zappa - Joes Garage comes out amazing with this procedure! Standard settings...

 

[himey]

Tool - 10,000 Days starts out a bit wonked on the first few tracks but makes an excellent comeback on the final eight tracks. I nearly shut it off and deleted, glad I didn't...

 

[himey]

Gong - Acid Motherhood Wow! Zero wonkiness on this one. Somehow a lot of the bass makes it over to the rears to go along with the panning effects. This one may be the best results yet of the 10-15 I have done.

 

[zeerround]

Getting close to the next release of SpecWeb.

I think I have a pretty good automatic fix of the rare spikes/transients that can occur in the rear channels. Some formulas for setting the various parameters of the 5.1 Master Limiter. Surround loudness it typically within 0.5dB of stereo across my test suite of almost 40 songs of all different types, styles, loudnesses, etc., and includes any "problem" tracks people have pointed out.

Also, during Beta testing, a few interesting things popped up re: inter sample or True Peaks. We found a track that had Ture Peaks above 0dB (you can use Adobe Audition Amplitude Statistics to detect these, but not Sound Forge) that caused glitches or spikes/clicking in SpecWeb.

A simple fix was just to lower the pre gain by 1 or 2 dB, but it sent me on an exploration of inter sample peaks and I'm wondering if those are the source of the above mentioned occasional transients.

We also discovered that different resamplers, used to upsample tracks before processing with SpecWeb, have different behaviors as to inter sample peaks. Some preserving them and others "fixing" them so that there are none above 0dB. That 2nd approach also changes the loudness and dynamics of the track, so I'm sanity checking how much of the "stereo to surround conversion reduces the loudness" is really due to the resampling process.

Fun fun fun...

Anyway I've taught SpecWeb to measure True Peaks and am experimenting with what to do with that info.

By the way, Re Sound Forge, that is normally my "gold standard" for all audio measurements, and I tend to trust it more than Audition (agrees exactly with SpecWeb RMS & Peak) but in this case it refuses to show True Peaks above 0dB. Sometimes you can still get a sense that one exists, when the Peak value is less than 0dB and the True Peak IS 0dB, but Audition shows a value above 0dB and agrees closely with the library I'm using in SpecWeb for loudness and True Peak.

 

[Plan9]

Getting close to the next release of SpecWeb.

I think I have a pretty good automatic fix of the rare spikes/transients that can occur in the rear channels. Some formulas for setting the various parameters of the 5.1 Master Limiter. Surround loudness it typically within 0.5dB of stereo across my test suite of almost 40 songs of all different types, styles, loudnesses, etc., and includes any "problem" tracks people have pointed out.

Also, during Beta testing, a few interesting things popped up re: inter sample or True Peaks. We found a track that had Ture Peaks above 0dB (you can use Adobe Audition Amplitude Statistics to detect these, but not Sound Forge) that caused glitches or spikes/clicking in SpecWeb.

A simple fix was just to lower the pre gain by 1 or 2 dB, but it sent me on an exploration of inter sample peaks and I'm wondering if those are the source of the above mentioned occasional transients.

We also discovered that different resamplers, used to upsample tracks before processing with SpecWeb, have different behaviors as to inter sample peaks. Some preserving them and others "fixing" them so that there are none above 0dB. That 2nd approach also changes the loudness and dynamics of the track, so I'm sanity checking how much of the "stereo to surround conversion reduces the loudness" is really due to the resampling process.

Fun fun fun...

Anyway I've taught SpecWeb to measure True Peaks and am experimenting with what to do with that info.

By the way, Re Sound Forge, that is normally my "gold standard" for all audio measurements, and I tend to trust it more than Audition (agrees exactly with SpecWeb RMS & Peak) but in this case it refuses to show True Peaks above 0dB. Sometimes you can still get a sense that one exists, when the Peak value is less than 0dB and the True Peak IS 0dB, but Audition shows a value above 0dB and agrees closely with the library I'm using in SpecWeb for loudness and True Peak.

Actually SoundForge shows True Peaks above 0dB if you work with a 32bit Float (or more) file. It will also indicate those TPs on lower bit depth files by showing +0dB in bright red (but won't actually tell you how far above 0 it goes). It's conservative that way...

Also I would be wary of any software resampler that limits the peaks after resampling. It would do more than what it is supposed to do and that's not good practice in audio.

 

[mlrocker]

SPECWEB, drag and drop eh? Cool, I'll try it.

 

[zeerround]

Actually SoundForge shows True Peaks above 0dB if you work with a 32bit Float (or more) file. It will also indicate those TPs on lower bit depth files by showing +0dB in bright red (but won't actually tell you how far above 0 it goes). It's conservative that way...

Also I would be wary of any software resampler that limits the peaks after resampling. It would do more than what it is supposed to do and that's not good practice in audio.

I don't see the red 0dB in sound forge. What version and where do you see that? In tools--> statistics?

Also, please tell me more about resamplers and "proper" settings for mastering. I think most people read the resampler documentation and automatically reach for the 2nd pass and normalize settings, "to avoid peaks above 0dB", but as you say this results in the dynamics/loudness of the track being changed.

I'm now thinking, after doing some more observation and testing with a pregain lowered SpecWeb, that (at least for purposes of SpecWeb) we would NOT want 2nd pass and normalization. We would want to preserve the dynamics of the original track, trusting in the original mastering engineer.

I guess it could be an option (avoiding intersample/true peaks above 0dB pre SpecWeb), but perhaps should NOT be the default?

Re flavors of free resamplers, we have r8Brain (free version), sox, the SSRC from 2008 in ea3to, and the modern SSRC, to name a few. Any opinion as to the sonically best and preferred settings?

For upsample I'm thinking no 2nd pass and no normalization, as I mention above). For down sample I don't use dither if keeping the same bit depth or the ending bit depth is 24 (24 bits is default output of SpecWeb, internally it is 32 bit float) but if going to 16bits what would be the "best/correct" dither settings for a preferred resampler?