SPECWEB (Now 2.2)

[EMR]

It’s a VMware Fusion virtual machine running on a Mac Pro 5,1 of 2010. The host operating system is the latest version of macOS Mojave.

EDIT: Although not officially supported, this computer is capable of running the latest version of macOS (10.15.x), named Catalina. It can also run natively any version of Windows 10 (both 32-bit and 64-bit).

 

[zeerround]

Edit: Just saw the "2010". Spec 2.0 requires a CPU with AVX instructions. That's any Intel or AMD CPU made since 2011.

The issue is likely support for AVX/AVX2 CPU instructions. This is working in Oracle VirtualBox. I haven't tried VMware Fusion. You should look for a setting that enables AVX support.​

​

I tried to compile a version of 2.0 for another member without AVX but AVX is now required in other libraries I'm using like current versions of Intel IPP and TBB, so we gave up and he is updating his computer.​

 

[zeerround]

https://www.pro-tools-expert.com/pr...eese-graters-used-in-todays-recording-studios

 

[EMR]

Thank you for your expert explanation. The instruction sysctl -a | grep machdep.cpu.features doesn't include AVX among the CPU features of my Mac Pro 5,1. I'm intrigued that VirtualBox sort of emulates AVX/AVX2 instructions. I don't suppose VMware Fusion will do that, but I'll try to find out. Many thanks.

 

[zeerround]

I did not mean to imply that virtualbox emulates AVX. I wrote that before I noticed the "2010" and was assuming you had a CPU that supported AVX but that VMware Fusion wasn't. I was indicating that Virtualbox (on a cpu with AVX support) has been tested with SpecWeb2.0.

Sorry...

 

[EMR]

Thank you for clarifying that. I know AVX/AVX2 are also available for Fusion if the host CPU supports that.

EDIT: I've just found out that there are versions of Intel® Software Development Emulator for macOS, Windows and Linux. In theory, you can run a non-compatible application (as far as one's CPU is concerned) from within an SDE shell, which, in this case, should provide the necessary AVX/AVX2 features, but I don't know if the execution will be fast enough to make it feasible. I'll try to do a little more research, but I'll only report back if a achieve some level of success. I'll try to install it on the Windows virtual machine itself, not macOS.

 

[J. PUPSTER]

Much thanks to Glenn (@zeerround) for helping me get going with this ! Much time and help was involved to get my puny brain to comprehend this stuff. It was mostly about configuring the computer (in my case a new NUC did the trick.)
Now on with some of those custom command lines...

 

[EMR]

I don't know if this is the right place to comment on the performance of version 2.0, but, for what it's worth, here goes. Months ago, I used SpecWeb 1.5 on a stereo version of Johann Strauss' famous waltz An der schönen, blauen Donau, Op. 314, which has a duration of 11 minutes and 50 seconds. The processing time was 3:30 minutes and most of the sound came from the front speakers, the surround effect being hardly noticeable. Yesterday, I used SpecWeb 2.0, with and without the -m4 parameter. The difference between both versions was hardly noticeable. The version generated using the -m4 parameter took 7:45 minutes to process. The one generated without said parameter took 8:32 minutes (go figure!). In both cases, the first pass took most of that time (roughly 6 minutes using the -m4 parameter, and 7:20 minutes the one without). I take it for granted that the regular AVX version of SpecWeb 2.0 will take considerably less time to process on a more modern, AVX-compliant processor similar to mine (a 3.06 GHz 6-core Intel Xeon).

The surround improvement relative to SpecWeb 1.5 is spectacular. In fact, most of the sound now comes from the rears. Naturally, this might be due to differences not only in the SpecWeb binary, but in the SpecWeb.ini file. So, it is likely that many classical stereo recordings may now be processed with greater success as far as creating a credible surround version is concerned. However, for what it's worth, I should point out that the most credible artificial surround version of said waltz I've heard was generated by treating the original stereo as if it was a UD-4 quadraphonic matrix, the process being done using an Adobe Audition 3.0.1 script on my Windows virtual machine. I don't recall how long that actually took to process. Compounding the script running time, and then, using Audacity for macOS, putting together the two stereo waveforms the script creates, the generation of a 5.0 FLAC (with a blank centre channel) must have taken roughly 15 minutes.

 

[zeerround]

From a PM with EMR:

Re: UD-4 quadraphonic matrix for stereo mixed tracks, if you are interested in exploiting phase information via SpecWeb you could experiment with the ambiance extraction settings.​

​

If you like the "I am in the audience" vs. "I am in the band" for classical that may suit your tastes more. Maybe use widths to put less stuff in the rears and add ambiance extraction to put decoherent (phase difference between L and R) sounds in the rear.​

and yes, modern machines run SpecWeb much faster, taking advantage of AVX instructions and newer versions of Intel performance libraries.

 

[deepsky4565]

I have been enjoying the new 2.0 version. It does a great job of making the surrounds sound discrete. One problem, which isn't Specweb's fault is that many stereo mixes don't mix across the stereo stage, with more things are far right and left, which then are put into the surround channels, with little left in the front right and left. I've taken files processed by Specweb 2.0, and used audio software to move the drums from the rears to the fronts to try to balance things out. The drums can be extracted with automated software, and are very common elements in pop/rock music. I've even bettered in my opinion a few inferior multichannel mixes. For an example, Paul McCartney's Mull of Kintyre on McCartney Years. It just doesn't sound "right" to my ears. Using the original stereo mix, processed with SpecWeb 2.0, it sounds "right" and is a very satisfying alternative. Well mixed stereo files are usually excellent contenders. I don't recommend stereo songs before the late 60's, since they, due to track numbers and mixing limitations usually end up being triangle mixes, with usually vocals in the center, and most everything else in the surrounds, similar but worse than my point above.

Some music you may not have noticed before have off center lead vocals, and it becomes painfully obvious with this method.

One issue on the software, is I've noticed sometimes the converted files will have at the end a repeat of the final few seconds of the song. I hope this is abug that can be corrected. Another minor thing is it will not take MP3s natively, you'll have to convert them first.

Big thanks to the creators of this new version, it is the most exciting thing to come along to me in a long time, well, other than the DV SACDs, The Beatles surround discs, and Alan Parsons' remixes, too.

PS, I finally read some of the documentation, and see a lot more about how it works, but most of my post here is still on the mark for the default settings. I can tell a lot of work went into this program!

 

[minimumtumbleweed]

Another minor thing is it will not take MP3s natively, you'll have to convert them first.

I would not convert mp3s to a lossless format for the purposes of conversion. There's a reason the software requires lossless format; it needs as much data as possible from the file to do its best estimates for channel separation. You might be able to fool the software into thinking the file is lossless, but it's not. You can't bring back anything that was lost in the compression to mp3, and you will most likely introduce artifacts into the track that will further complicate your attempt to convert to surround. You're better off finding the lossless source, such a CD or DVD and ripping straight from there, or downloading in FLAC from online stores that allow such formats.

 

[deepsky4565]

Yes, I know, but when I have no other copy of a song, that's what I'll do. Just pointing that out.

 

[zeerround]

Yeah besides the quality issue, the audio file library I'm using, libsndfile, doesn't support mp3 due to licensing restrictions, and because of the quality of lossy conversions I haven't thought about adding support for mp3.

 

[zeerround]

So hey, I figure we could all use a new toy or distraction in these social distancing times, so posted a link to SpecWeb 2.1 at the top of this thread.

Besides a few bug fixes the main changes between 2.1 and 2.0 alpha/beta releases are the default settings.

I changed a lot of things, backing out of my "purist and mathematical" approach and back to a more subjective listening approach. Widths, Levels, and Default ArcTan steering methods have all changed. In the ini file I documented what those setting were in 2.0 and 1.5, should you not be happy with the latest or just want to further understand the changes.

FYI, the subjective listening was based on over 50 example tracks of all genres and production types, using a BBC studio virtual listening room with personally calibrated headphones on a Smyth Research A16.

Zone 2 is the new default ArcTan mode. Zone two being preferred of other options by users.
Note: The SpecWeb Play web interface has not been updated for Zone 2 yet. That will be addressed in SpecWeb 2.2. Version checking is also not working.

Bug Fixes:

Echo at end of some files that fade to silence (when using VST)
Tagging a mixture of tracks with tags and no tags did the wrong thing

So check it out and let me know what you think. I'll keep working on stuff like updating the Web Interface for Zone 2, etc. in the mean time.

Stay Healthy,
Glenn

 

[minimumtumbleweed]

I've converted a lot of my library to surround using SpecWeb v1.5 (thanks!). I've tried doing some of my files over with SpecWeb 2.1, but a number of files that I had no problem converting before now give me the error:

Launch-VST Process failed with Status Code C0000005

It then goes through this 20 times and ultimately fails to convert. Anyone else experiencing this?

 

[zeerround]

Does it happen on every song, or only some?

What happens if you add -v1 (show the VST)?

 

[minimumtumbleweed]

Does it happen on every song, or only some?

What happens if you add -v1 (show the VST)?

Not every song, but about 40 out of the 300 I tried to convert. But they always fail, and did not fail in 1.5. Using the -v1 flag opens up the VST, which has lots of bars that jump around, but it still fails. Can't really pinpoint what the hangup with those specific files are.

 

[zeerround]

Interesting. maybe you could PM me with a link to one that fails?

 

[zeerround]

OK this is a confirmed but. Very odd... but working on it.

 

[deepsky4565]

I really enjoyed 2.0a, however there were a few of the couple thousand tracks I converted that sounded "lousy." Hard to explain, but some sounds sounded harsh, and somewhat distorted. Enya tracks Amarantine and How Can I Keep From Singing? were 2 particulars. Good news is that this 2.1 sound much, much better. Thank you so much! I'll be making a donation when we're through this financially tough time.

Also as I mentioned earlier in this thread, I also do work to upmix mono to stereo, but using those tools, I've also been able to greatly improve some of my upmixes, and it is very satisfying!