In general, does anyone know if there is a number of votes a feature request needs to get noticed?
There is no set number, whatever works best with their updates and or roadmap etc.
2 Likes
From my very loose knowledge of machine learning (please correct me if Iām wrong); they wonāt be able to āupdateā the current algorithm. Inprovements would basically have to come from retraining for longer, on a larger, better quality data set.
Iād love it if they could, though. Iāve made a few captures in the ~year of ownership. Thereās always some subtle inconsistencies with the mids that I havenāt been able to EQ out. I need to feed it into the comparative EQ thing in logic and see what it spits outā¦
1 Like
I also only use on my desk but I want the desktop editor so I can put it on the floor. For example using the looper can be really good for writing riffs and seeing if things fit but controlling it with your hands simply sucks for that application.
2 Likes
Iām also not versed in it so take it with a pinch of salt but I would think there is a limitation with the cpu thatās in the box and more accuracy requires more cpu, even if they did update the algorithm. This is why NAM has an advantage because you can use high powered computers to do the work and essentially run much more cpu/gpu intensive processes. Having said that, I also think the capture accuracy reaches a point where for most people itās accurate enough or the aliasing isnāt noticeable and overall it becomes a bit like a snobbery contest as to how accurate things are when most listeners wonāt know or care anyway. People tend to be more obsessed with knowing things are accurate rather than deciding if they actually like the sound or not. The capturing process should be close enough but I think most people just donāt dial the settings properly before capturing and it comes out like crap. Also, most of the ones you can buy on the market tend to be better because theyāre using a much better signal for the capture and have more experience doing it with other profilers previously.
1 Like
I would love to see a standard emerge for captures that works on any or most high-quality modelers - across brands. Letās decouple the capture process from the modeler. Not to say that I wouldnāt keep the ability to capture on the modeler, but letās make the capture file format agnostic so the capture can be ported/downloaded from ābig hardwareā to any modeler if desired.
The hardware included on modelers will probably never be able to compete with what a beefy cloud server can render or even what can be generated when on your home PC by virtue of upgraded memory, GPU, and general processing power, as well as the ability to generate long duration captures in the background while you continue to use your modeler.
The hardware on your modeler also tends to be static as opposed to the upgradeability of a PC or server. The hardware on the device at the time of purchase is what you are stuck with.
The cork-sniffing debate also has some legitimacy, as you may get to a point with captures where you hit the point of diminishing returns and the differences in the āqualityā of the capture become inaudible, and strictly statistical.
I enthusiastically support the ongoing efforts to enhance the onboard capture results for the QC. In the big picture though it just seems like some degree of capture format standardization is the path forward for leveraging hardware for capturing that canāt be placed in a floor modeler due to price and footprint/size limitations.
1 Like
I donāt do captures and probably never will. I am the type that likes to use presets and models that are already there or built by people that know more about it than I do. What Neural will have the ability to do regardless is make the Models that are in the QC better as time goes by using the resources they have there and THEN give us timely upgrades to existing models of amps and other effects, as well as adding new ones. We may very well be limited on our end but their capacity to make improvements on their end is most likely not. It is just a matter of what priority they give any specific upgrade or improvement.
3 Likes
Interesting, I went from the QC to a pedalboard with the Dream 65 and now back to QC. I was struggling to get the QC to sound like the Dream 65 but a little 9 band eq and York Audio cab addition and I have a sound I like better than the Dream 65. The one thing that really helped the QC was running it into my Walrus Canvas di. It really gave it the clarity that the Dream 65 had (which I was also running into the Canvas as it has no XLR output). In the end I really liked having the routing options available on the QC for Sundays as well as adding cabs (I really donāt care for the cabs on the Dream 65, I only liked one out of the 6 available). At any rate, you have the right idea - keep both! Thanks for serving your church so faithfully, blessings brother!
2 Likes
Hi Scott and thanks for sharing! Yes it is quite an adventure. I could have saved a few dollars by going a different way but in the end it is all worth it! Yes, playing on the worship team with my good friends every week is always a joy and a pleasure. I have been doing it for 14 years and it has made me a better guitarist and musician. This Sunday I get to play and then fill in for my pastor who is out of town! I am going to speak on āFishing for People.ā God bless!
Oh, and yes, the routing options on the QC are great. Which amp do you like best on the QC? Do you use a modeled one or a capture? I had been using the Morgan 50 from the QC with the Keeley AT Super Mod overdrive pedal in front of it and then the Keeley Halo and Strymon Cloudburst into the effects loop. Previously I had been using a Preset from Worship Tutorials of a 64 Bassman that sounded really nice. The QC has so much potential and will continue to get better and better!
1 Like
Outstanding! I really need to explore the Morgan, but Iāve been just using the modeled Fender Deluxe (although I also really like the Tone Junkie captures of the deluxe also). I was really inspired by Justin Ostrander on YT when he did a video on the Fender Deluxe. He has a great channel, itās worth a watch!
Very cool about the Halo and Cloudburst! I just saw that Strymon released a new āBrigā pedal in the same size as the cloudburst. If they make a Flint or El Cap I will add that to my board for sure! I have my QC mounted on a Creation Elevation board so I have room for a few pedals.
1 Like
I seen that Brig pedal for a few minutes on Andertons yesterdayā¦ looks good. I am not sure if I have seen Justin Ostrander but will check him out. Great to connect with you here! Haha, I just checked and I do follow him already. He is very knowledgeable.
1 Like
The QC with its old/cheap SHARK CPU isnāt going to compete with NVIDIA Tensor core equiped GPUās or Appleās cutting edge M line of chips in regards to neural network computing.
Best we can hope for is that Neural releases their modified NAM PC/MAC Plugin that creates captures that are usable on the QC. And that they update the captures of the devices on the the Unit by themselfs.
Captures done with the QC will stay the way they are, imo the underlying components arenāt flexible enough to make this work.
Training the neural network is what takes all the power. Once itās trained, it takes exponentially less processing power to apply that output algorithm to an input - like the one it already runs. Thereās no reason why it couldnāt run an updated one.
2 Likes
But you are training the NN for every individual capture, thatās why precision capturing on NAM takes hours on beefy hardware.
Just using an existing algorithm to approximate the desired output (aka gain matching) is believed to be what Kemper is doing. And has nothing really to do with neural networks at that stage anymore.
I feel like thereās a misunderstanding here.
Using your video at ~4m25s, you can see a basic outline of how these things work.
That āHidden Layerā is the portion that has had to be generated via deep learning on a large set of training data. NDSP have their āHidden Layerā, NAM has theirs. I assume the QC runs a set number of epochs, whereas NAM is configurable (hence the ability to spend days taking a capture, although dude proved itās pointless, but good to know).
What Iām saying is Iād hope that NDSP are able to improve upon their dataset, buy more time on a supercomputer, and maybe churn out a more accurate algorithm. I know itās complicated af though, for all we know theyāre trying, but it keeps spitting out worse results.
It would be nice a user menu selectionā¦
Several levels of quality (and time).
Then we can decideā¦
The video linked above showed that gains from more time diminish pretty quickly.
At 1000 epochs, start of diminishing returns, heās taking 60min per capture on a 4090 gfx card. QC hardware will never be able to match that or even attempt this kind of computing. He also showed how much farther along NAM is compared to the QC captures. And you can see how much the quality raises by just upping the epoches to 300 for exampleā¦
A NAM to QC plug in conversion tool from Neural would be really fucking great and make the QC by far the best NAM loader in town. Tone X would pale in comparison ā¦
Guys, do you have noticed that NAM has a lower S/N ratio?
With several audio interfaces (but no problem with commercial plugins and so onā¦) in my case the S/N ratio itās not among the bestā¦ not dramaticā¦ but the captures of tonex or qc are really silent.
QC isnāt as powerful as 4090?
NO SHIT lol
NAM compatibility would be sick. That has given me an idea though: outsourcing the capture process to desktop machines via the desktop editorā¦