|
Post by superwack on Dec 5, 2020 19:38:46 GMT -6
All my testing included comparing a final 44.1 down-sample. Even MP3s sounded better when the original recording and all the processing was done at 96k. It meant buying a new computer, so my prejudice was against it. I'm sure your testing was 100% more scientific than mine but, to be honest, I'm not sure mine could have been. 2 years ago or so some colleagues and I were discussing this and we did the following: Downloaded FREE Multitrack stems from one of those websites so we had no bias about who had done what on what track. We were unconcerned with the material other than one had to be 44.1k/24 bit and another 96k/24 (I do remember us trying to keep the track count manageable like maybe 20-30 tracks) We "mixed" the songs with static levels, pans and added a ridiculous number of harmonic plugins. There was NLS and VTM and decapitator and saturn and various modeled this and thats driven slightly in to saturation - they were set to "do" something we were not trying to make a masterpiece. we mixed down the 44.1 and imported the mix (all individual tracks and plugins) in to a 96k session then bounced that down and imported the stem back in to the 44.1 and vice versa (so we had 1 session with native 44.1 and 44.1 downsampled from 96) and a session with the upsampled 96 and imported/upsampled the 44 in to the 96 we did the same with the native 96k So we wound up with: native 44.1 bounce vs. downsampled 96 (that originated at 44.1) bounce upsampled 44.1 bounce in the upsampled 96 session (bounced) native 96 bounce with upsampled 44.1 (that originated as 96) bounce downsampled 96 bounce vs downsampled 44.1 bounce (if I screwed that up read it as "every combination of native and upsampled/downsampled - you can't mix sample rates in Pro Tools so it was the best we came up with) In every single case myself and my 3 colleagues all picked the 96k versions in blind test as sounding ever so slightly "better" slightly more open, a tiny bit less crunchy. Even the 44.1 that was upsampled to 96k then bounced down and re-imported in to the 44.1 sounded slightly better. Obviously there weren't huge differences but maybe a lack of aliasing, or cramped curves or something else. who knows. maybe it doesn’t matter but we could for sure hear it.
|
|
|
Post by chessparov on Dec 5, 2020 22:54:53 GMT -6
So 44.1 upsampled to 96K... Is the new Pro Tools into 2"? Chris
|
|
|
Post by superwack on Dec 5, 2020 23:49:29 GMT -6
So 44.1 upsampled to 96K... Is the new Pro Tools into 2"? Chris Yes or maybe to 1/4” 🤔 if there is any “magic” it’s the plugins working at 96 as upsampling adds nothing to the audio except some sweet, sweet zeros 😂
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Dec 6, 2020 11:16:57 GMT -6
All my testing included comparing a final 44.1 down-sample. Even MP3s sounded better when the original recording and all the processing was done at 96k. It meant buying a new computer, so my prejudice was against it. I'm sure your testing was 100% more scientific than mine but, to be honest, I'm not sure mine could have been. 2 years ago or so some colleagues and I were discussing this and we did the following: Downloaded FREE Multitrack stems from one of those websites so we had no bias about who had done what on what track. We were unconcerned with the material other than one had to be 44.1k/24 bit and another 96k/24 (I do remember us trying to keep the track count manageable like maybe 20-30 tracks) We "mixed" the songs with static levels, pans and added a ridiculous number of harmonic plugins. There was NLS and VTM and decapitator and saturn and various modeled this and thats driven slightly in to saturation - they were set to "do" something we were not trying to make a masterpiece. we mixed down the 44.1 and imported the mix (all individual tracks and plugins) in to a 96k session then bounced that down and imported the stem back in to the 44.1 and vice versa (so we had 1 session with native 44.1 and 44.1 downsampled from 96) and a session with the upsampled 96 and imported/upsampled the 44 in to the 96 we did the same with the native 96k So we wound up with: native 44.1 bounce vs. downsampled 96 (that originated at 44.1) bounce upsampled 44.1 bounce in the upsampled 96 session (bounced) native 96 bounce with upsampled 44.1 (that originated as 96) bounce downsampled 96 bounce vs downsampled 44.1 bounce (if I screwed that up read it as "every combination of native and upsampled/downsampled - you can't mix sample rates in Pro Tools so it was the best we came up with) In every single case myself and my 3 colleagues all picked the 96k versions in blind test as sounding ever so slightly "better" slightly more open, a tiny bit less crunchy. Even the 44.1 that was upsampled to 96k then bounced down and re-imported in to the 44.1 sounded slightly better. Obviously there weren't huge differences but maybe a lack of aliasing, or cramped curves or something else. who knows. maybe it doesn’t matter but we could for sure hear it. It’s all shitty plugins. Waves, Slate, Sound Toys all alias horribly and have defective dynamic processing. Higher sampling rates won’t save the audible band from compounding intermodulation distortion from aliasing either. You need anti-alias filters in all non-linear processes. That’s what the steep cut is in the UAD plugs. The Tokyo Dawn guys are right in spite of the entire post 1996 or so pop music industry.
|
|
|
Post by superwack on Dec 6, 2020 12:30:55 GMT -6
I'm sure your testing was 100% more scientific than mine but, to be honest, I'm not sure mine could have been. 2 years ago or so some colleagues and I were discussing this and we did the following: Downloaded FREE Multitrack stems from one of those websites so we had no bias about who had done what on what track. We were unconcerned with the material other than one had to be 44.1k/24 bit and another 96k/24 (I do remember us trying to keep the track count manageable like maybe 20-30 tracks) We "mixed" the songs with static levels, pans and added a ridiculous number of harmonic plugins. There was NLS and VTM and decapitator and saturn and various modeled this and thats driven slightly in to saturation - they were set to "do" something we were not trying to make a masterpiece. we mixed down the 44.1 and imported the mix (all individual tracks and plugins) in to a 96k session then bounced that down and imported the stem back in to the 44.1 and vice versa (so we had 1 session with native 44.1 and 44.1 downsampled from 96) and a session with the upsampled 96 and imported/upsampled the 44 in to the 96 we did the same with the native 96k So we wound up with: native 44.1 bounce vs. downsampled 96 (that originated at 44.1) bounce upsampled 44.1 bounce in the upsampled 96 session (bounced) native 96 bounce with upsampled 44.1 (that originated as 96) bounce downsampled 96 bounce vs downsampled 44.1 bounce (if I screwed that up read it as "every combination of native and upsampled/downsampled - you can't mix sample rates in Pro Tools so it was the best we came up with) In every single case myself and my 3 colleagues all picked the 96k versions in blind test as sounding ever so slightly "better" slightly more open, a tiny bit less crunchy. Even the 44.1 that was upsampled to 96k then bounced down and re-imported in to the 44.1 sounded slightly better. Obviously there weren't huge differences but maybe a lack of aliasing, or cramped curves or something else. who knows. maybe it doesn’t matter but we could for sure hear it. It’s all shitty plugins. Waves, Slate, Sound Toys all alias horribly and have defective dynamic processing. Higher sampling rates won’t save the audible band from compounding intermodulation distortion from aliasing either. You need anti-alias filters in all non-linear processes. That’s what the steep cut is in the UAD plugs. The Tokyo Dawn guys are right in spite of the entire post 1996 or so pop music industry. It’s so weird/2020 that people are alias deniers. You can see it, measure it and hear it. When aliasing analysis started to become a thing on GS I remember testing NLS and kept hearing this sound that sounded like a radio being tuned in the background. Not fully understanding I put in a call to Waves and after a bit of back and forth on the phone then subsequent emails I came to understand that Waves do not think this is a thing that needs to be addressed or even acknowledged which explains while almost all of their plugins exhibit this issue. I get the argument that it’s not “real world” to have 0dbFs at 20k but there are plenty of plugins that don’t have any aliasing and therefore none should. My 2c
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Dec 6, 2020 14:57:15 GMT -6
It’s all shitty plugins. Waves, Slate, Sound Toys all alias horribly and have defective dynamic processing. Higher sampling rates won’t save the audible band from compounding intermodulation distortion from aliasing either. You need anti-alias filters in all non-linear processes. That’s what the steep cut is in the UAD plugs. The Tokyo Dawn guys are right in spite of the entire post 1996 or so pop music industry. It’s so weird/2020 that people are alias deniers. You can see it, measure it and hear it. When aliasing analysis started to become a thing on GS I remember testing NLS and kept hearing this sound that sounded like a radio being tuned in the background. Not fully understanding I put in a call to Waves and after a bit of back and forth on the phone then subsequent emails I came to understand that Waves do not think this is a thing that needs to be addressed or even acknowledged which explains while almost all of their plugins exhibit this issue. I get the argument that it’s not “real world” to have 0dbFs at 20k but there are plenty of plugins that don’t have any aliasing and therefore none should. My 2c “Fred” and some other members are not measuring foldback aliasing on gearslutz. They’re measuring imd from clipping and other non-linear processes and the imd from aliasing is going to be beneath their extreme imd from being totally incompetent. Many pieces of analog gear have spurs lower in frequency than the fundamental from compounding non-linear processing anyway. The Hammerstein graph in plugin doctor is better for revealing aliasing anyway. There are plugins with aliasing in the audio path that beat plugins without it. The PSP Infinistrip gets a little farty, boxy, and 2d but beats the Slate and IK racks and half the UAD plugs any day. It’s repackaged old school dsp that still works. The Waves Renaissance plugs are still good. The real problem is the build of distortion in the treble, build up of the slight aliasing in the midrange, and dysfunctional plugins. If you don’t abuse the drive and high shelves in many plugins, you’re not going to hear it anyway with the color plugs gainstaged well. The problem is the stack up from waves and sound toys plugs gets really bad when they’re pushed so that you can hear the distortion. Using them for parallel sends is nutty too because you just have a layer of aliased grind under it. It’s also impossible to have zero latency non-linear processing without aliasing or phase shift as bad as a cheap 1990s converter. This is where Infinistrip is great and Scheps Omni Channel is okay. The problem is their compressors basically don’t work as advertised and the Scheps FET doesn’t work at all. Low latency is where higher sample rates excel. But you still need something like 188.2/192khz minimum for a softer knee, fast attack compressor to work right and maybe 352.8 to 768 kHz or even greater a fast attack harder knee to work right. You can’t emulate many tape machines without at least 352/384 kHz For the bias tones. This is where many plugins like the DMG and Pulsar collapse. They don’t work correctly until they oversample massively, sometimes don’t work at all, and their oversampling is wildly inefficient for any current computer in a real session. 8x oversampling sometimes takes the Pulsar compressors 40x the cpu. You can be stuck with a DMG Trackcomp 2 1176 with 4x oversampling, no aliasing in the audio, yet it’s still clicking like hell and you’d be better off with a Waves L2/ Rcomp or the limiter in Infinistrip, aliasing in audio path be damned. This is why the Tokyo Dawn plugs and most of Ray Dratwa’s plugs are so cool. They upsample only when needed and prevent most user errors and guys who don’t think they have to enable oversampling because they bought a 500 dollar cpu and run at 192khz or have an ancient laptop.
|
|
|
Post by chessparov on Dec 6, 2020 15:57:35 GMT -6
Realistically, just being at the "Demo level" here then... Along with the PSP line, are there a couple other Plugin lines, you guys recommend. (I probably/eventually will get at least one Valhalla reverb BTW)
Thanks, Chris
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Dec 6, 2020 16:30:34 GMT -6
Realistically, just being at the "Demo level" here then... Along with the PSP line, are there a couple other Plugin lines, you guys recommend. (I probably/eventually will get at least one Valhalla reverb BTW) Thanks, Chris You can do it for free. Molotok if you like the sound and Dragonfly verb and the eqs in your daw. Slick eq for some free vibe. demo you might get paid for? Easy vocal compression for cheap? The Purple MC77 to easily take the peaks off despite the bright sound and the Fuse VCL-4 for leveling. Only buy the purple for 30 bucks or so on sale. Otherwise, just use a limiter. I like Goodhertz Faraday’s tone but it’s 100 dollars. Molot can do it too but is a bit harder to set up than an 1176 and an optical compressor. Klanghelm DC8C can do it all for 20 bucks but Molotok is FREE! Try it. Reverb is stylistic. Valhalla Vintage is decent but not the best you can do. It’s super easy to use and cheap and hard to make sound like a trash can. The Rare Signals plate sounds awesome but is expensive. The Kush / Relab plate is cool too and a bit more lofi and adjustable. PSP 2445 rules, is easy to set up for a digital reverb, and sounds great. UAD AMS RSX 16 for that 80s vibe. Lexicons can get weird. Not a huge fan. Epicverb and Dragonfly reverb are still FREE!
|
|
|
Post by chessparov on Dec 6, 2020 16:55:56 GMT -6
Thanks! BTW I mean "Demo", in the sense of "practice recordings", in order to hear what I might be doing well, and what can use improvement. Besides a "reveraholic" like me, can't be totally trusted! Also to compensate for the Comatose state of performing around Town. Learn new songs too, etc. Chris
|
|
|
Post by superwack on Dec 6, 2020 21:23:30 GMT -6
<snip> The Hammerstein graph in plugin doctor is better for revealing aliasing anyway.
Would you mind giving some quick tips on how to read a Hammerstein graph / what to look for? I tried to read the linked paper from the Plugin Doctor manual and sorta, kinda understood it but the examples shown in that paper give a theoretical vs. actual plot for each model which made more sense to me than how it is displayed in PluginDoctor. I couldn't find any help on google, youtube, etc. I think I get what it's doing but not what's good/bad. Looking at the PRE-73 it has relatively flat lines which I'm assuming is "good" based on you liking that plugin (thanks by the way, bought it after demoing after you talked about it) but there are other ones where the lines are all over the place. Appreciate any help, Thanks.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Dec 6, 2020 22:24:35 GMT -6
<snip> The Hammerstein graph in plugin doctor is better for revealing aliasing anyway.
Would you mind giving some quick tips on how to read a Hammerstein graph / what to look for? I tried to read the linked paper from the Plugin Doctor manual and sorta, kinda understood it but the examples shown in that paper give a theoretical vs. actual plot for each model which made more sense to me than how it is displayed in PluginDoctor. I couldn't find any help on google, youtube, etc. I think I get what it's doing but not what's good/bad. Looking at the PRE-73 it has relatively flat lines which I'm assuming is "good" based on you liking that plugin (thanks by the way, bought it after demoing after you talked about it) but there are other ones where the lines are all over the place. Appreciate any help, Thanks. The harmonics fold back. This is the FETpressor in Infinistrip. And this is a good plugin for tracking through and live use. Zero latency, low cpu, and nowhere near "crunchy" sonically. It beats a lot of oversampled plugins despite being a bit 2d/boxy due to the aliasing. Attachments:
|
|