Polyphonic Guitar to MIDI VST/AU "MIDI Guitar"- BETA TEST

VST, AU, AAX, CLAP, etc. Plugin Virtual Instruments Discussion
Post Reply New Topic
RELATED
PRODUCTS

Post

Yep, this is not good. We're taking leaps backwards with every new version.

I'm using it as a plug-in in Reaper (my ASIO latency is 96 samples) and driving Poly-Ana with it on another track. Hard to say for sure but latency doesn't feel any better or worse than previous versions.

Tuner is WAY off. (And some gradations to show cents between the notes would be nice.) It's looking as much as a fifth to even a quarter of the way to the next highest note sharper than it really is. (Which should be about 20 to 25 cents -- a LOT!)

Also, when all 6 strings are strummed open (the way you're supposed to use poly-tuners) it's only catching a few of the notes, and jumping around a lot. Tracking is WAY worse than previous versions, even version 0.6. Lots of octave jumps, missed notes, and double-triggering.

Changing guitar type to "Electric Guitar, Bright" and "Electric Guitar, Dark" both seem to help a bit, but not much, over "Electric Guitar, Default". (I'm using a very nice American Telecaster now, it should be about a best case signal. Using mostly the bridge pickup but trying others as well.)

I don't understand what Sustain "Up To Instrument" means. There shouldn't be any sustain unless I'm merging a pedal from another controller, right?

Pitch bend doesn't seem very sensitive at low bend amounts (as in vibrato, where it's most useful). It seems more sensitive to small amounts of bend when the pitch bend range is set low vs. high. It shouldn't. MIDI pitch bend resolution is more than sufficient so that even at an octave range very slight bends of only a few cents can still be expressed. (Is it doing something to "snap" pitches that are close to notes onto the note? If so, that would certainly work to defeat small bends and I'd recommend against it.)

And still no multi-channel? What's the point of having pitchbend at all then?

Pitch bend range control doesn't match between the Essentials and Advanced views. I assume they're meant to be the same control? It seems to take the setting of whichever one I moved last.

Sensitivity affects the waveform display, so I guess that's just a pre-gain before the analyzer? It used to have a limiter/compressor-like behavior but that seems completely gone now. (Maybe that's necessary for accurate velocity tracking?) Does it clip internally and create false harmonics when the waveform graph goes off the scale? I'm guessing it does as the tracking goes crazy when the waveform graph "clips". Is that necessary in floating point? Can't your analyzer deal with ranges over 0 dB or does something about the math break? (I'm definitely not inputting a clipped signal, if clipping is happening it's between the effect input and your analysis stage.) And if internal clipping IS necessary and always has this bad effect, maybe a big red light to indicate it would help.

Generally, higher sensitivity and lower pitch prediction settings seem to work better for me. (That pitch prediction control could really have more meaningful values than 1 to 3. Did you say they affect latency? Maybe it should say so?) But lots of false octave notes at higher sensitivities.

It's using a LOT more CPU now. I can't play a full 6 string chord (with Poly-Ana on default settings) without breaking up (which is odd as Reaper only shows it using 6% to 8% CPU and Poly-Ana less than half that.)

It's also very difficult to make it play a 6 note chord or even just 6 open strings. It seems to have trouble picking higher strings out when low strings are sounding, and vice-versa. Same with strumming an already-sounding string. It often misses the new attack unless you play stacatto (i.e. with silent spaces between notes). The combination of these two effects makes strumming chords virtually impossible. (Previous versions handled strumming chords quite well.)

One thing that does continue to amaze me is its release sensitivity. Even after my ear can't hear the string going anymore, the synth voice still sounds. Then I palm the apparently silent string and it magically cuts the note off. I'm surprised a single coil electric guitar has the dynamic range to do that. It's doing a better job of determining if the note is still playing than my ear!

Dynamics modulation doesn't seem to be doing anything. I'm playing the guitar's volume control quickly up and down after I strum and not hearing any effect of Expression CC #11 as a Volume Control. Same thing with the Brightness Controls not responding to my rapid movements of the tone knob. (Are these meant to track continuously, or does it only determine a value for them at the note-on time?)

After tweaking all the settings it's not quite so bad now. But I still think it's very unusual and ambiguous to have pitch bend on a single polyphonic channel (in monophonic mode, sure, that's fine). Where's the multichannel output? It should only be a few lines of code to implement. (Just give each note its own channel. Determining which string is which is MUCH harder, but when you're playing the same synth sound for all strings, who cares? You can improve on that later.)

:shrug:

Post

AdmiralQuality, I don't think it works as well with single coils. Have you tried a compressor or booster before the input?

I got incredible results using a strat-style Brian Moore guitar and the Seymour Duncan bridge humbucker. Jacked the 'sensitivity' way up and the velocity gain too and it was practically perfect. Absolutely amazing. I won't be buying a Triple Play anymore it was that good.

However with my P90 equipped Les Paul the results were significantly less good. Still good, but not stellar. Most notably, alternating on higher strings would often lose one string (entirely unlike with the humbucking guitar). A compressor increased sensitivity on higher strings but at the cost of the lower ones. I'll be experimenting with a hardware booster or comp on the input, but the humbucking guitar worked so well that it's just for experimentation's sake. I think it just prefers humbuckers.

Overall I feel 0.7 is significantly better than 0.6 (minus the scan bug, which sucks). I didn't use 0.6 as extensively but the overall 'feel' of 0.7 is better to me. As long as I use the humbucker guitar, that is.

Post

wasi wrote:AdmiralQuality, I don't think it works as well with single coils. Have you tried a compressor or booster before the input?
You're probably just a humbucking guy. I'm a single coil guy, can't stand humbuckers. But I'll try it with my old Squier Strat with Fender Vintage Noiseless pickups, which is one of the guitars I tested with before, and get back to you. (And I have another single coil Strat-like guitar that I also tested with before, with much better results. I'm not the only person here reporting it's gotten worse.)

And no, I wouldn't put any processing before it. If it needs some kind of processing or conditioning it should be built into the product. (Compression is just going to mess with velocity detection anyway.) It's running into a high impedance guitar input on a digitizing pre-amp, and has fantastic tone.

Post

I'm a P90 guy actually. :P

But the P90s didn't do so well. I'm glad I left the Humbucker in the bridge on the Brian Moore.

And come on, we put boosters in front of $3000 amps for better results, surely it's OK to put some on front of $100 software. ;)

Post

Trying it now...

On reloading the project I see it reactivated the Test Piano, even though I had it off when I saved the project.

Hmmm... it's different. But not different good, just behaving strange in a slightly different way. Less octave jumps now but still much worse tracking than 0.5 and earlier versions.

I noticed the tuner seems to be more off for the low pitches than the high pitches.

I wonder if it's better to tune to the correct pitch, or to what MIDI-Guitar thinks is correct. :\

I just verified the tuner issue by running Poly-Ana into it (which I KNOW is producing correct pitches) and the same thing. About 1/5 of a semitone sharp on the low E, ranging to maybe about 1/8 of a semitone sharp on the high E. Again I wonder if that tuning discrepancy is shared by the note determination logic.

Noticing more harmonic errors now. Also it's producing more than 6 notes! (Is this a new feature for 7 string guitars, or whatever? I don't think I'm seeing more than 7 at once... hard to say as it's so unstable they pop on and off while the same chord is playing.)

:(

[EDIT: This round was with the Squier Strat with Fender Vintage Noiseless pickups in the bridge position. Same guitar that I recorded and sent to Jam Origin in my earlier tests.]
Last edited by AdmiralQuality on Tue May 21, 2013 8:30 am, edited 1 time in total.

Post

wasi wrote: And come on, we put boosters in front of $3000 amps for better results, surely it's OK to put some on front of $100 software. ;)
I don't. (Well, I use a fuzz pedal, but obviously not in this application.)

My guitar signals are peaking just below 0 dB at the input. There's no need for any boost. (And again, if it needs some EQ like a mid boost, that should be built into the product.)

Post

Thanks everyone for all the feedback!

We have fixed the plugin scanner and indeed there was a typo in the tuner which made it tune way off sharp, mostly notable on low strings. This might explain why some have had tracking inferior to 0.6 if you have been using MG 0.7's own tuner.

Il'l just get back to you with answers after building 0.7.2 later today.

One question we get a lot:
We haven't actually officially released 0.7 on the website because we needed some hands on testing first to ensure that most users have a good experience with the upgrade. The plugin scanner and tuner are good examples of how things sometimes go wrong.

We'll be back soon.
JamOrigin.com

Like us on Facebook.com/JamOrigin and follow us on Twitter @JamOrigin

Post

There are currently two flaws which prevent the MG from being a final solution for me:

1. Simultaneous notes don't track well. It's more designed towards strumming than fingerpicking, but I don't want to strum a piano?

2. Chords with notes a semitone apart are a problem. The first note gets overridden.

Hopefully these issues can be addressed in future releases.

Post

JamOrigin wrote:This might explain why some have had tracking inferior to 0.6 if you have been using MG 0.7's own tuner.
I was tuning to a correct reference.

If I remember correctly (it's been a while) 0.6 had unsatisfactory tracking as well. I think 0.5 was the last one that was actually fun to play with. These last couple of versions have felt like steps backwards from there.

Post

Ok, just let me catch up on the latest posts.
Obviously a lot of scanner issues. These should be fixed in 0.7.2 which is available now (link in my next post).
We also put back the load single plugin function as requested. The tuning problem should be fixed too, although the poly tracking of the tuner is still considered experimental.
StickyWicket wrote:
Daggilarr wrote:Would it not be possible to have a system that first lists all plugins and gives a check or uncheck option of which to load. I have many plugins that are of no relevance to midi guitar. I realise this may not resolve this problem but would be a useful feature.
+1

I also like the more simple one of manually choosing which VST to use (i.e. V.6.1).

V.7.1 is more responsive and the feel is so much better (well done!!), but unfortunately, I can't access any of my VST's due to this scan issue. I'm using the Win32 and so have only tried it in standalone. I'm looking forward to the next scan/fix update. :)
Oh thats exactly with 0.7 does (?) - did you see the purple toggle buttons?
Please look at 0.7.2 now with scanner issues fixed and let me know if we can improve it further.


ariaJazz, karozas,

Hope your scanning issues are fixed with 0.7.2.
I've yet to experience stuck notes myself. It must have been loosing some audio frames - maybe you are cpu limited?
0.7 is certainly harder on the cpu - but i think we can optimize it in a 0.7.3 update.
I put input channels readability on todo-list.

AdmiralQuality,

Tuner should be fixed now. Obviously bad recognition will cause the tuner to jump around. More on this below.
No reason to go that low in buffersize. Optimal latency you get currently at 128 @ 44.1Khz. If its too hard on the system then go with 256 for now.
Multichannel comming. Most changes in 0.7 and also the cpu overhead you experience is actually related to our preparation for multichannel midi. Maybe we gave it too long a leash and so there are some optimizations left for us to improve on.
We are reimplementing the waveform display to proper db scale at some point. Just down-prioritized everything that is purely visual still.

The last few days of feedback on 0.7 have been quite surprising for us and hard to draw conclusions. Many feel 0.7 is a great improvement over 0.6 and others feel its significantly worse. Results most certainly has to do with playing style and type of guitar. 0.6 was very robust for most. This isnt very satisfactory, but we got some invaluable knowledge in the process.

We have a few more things to try out in the search for the "silver-bullet". If it won't success we will just bring in the sliders, compressors and presets you need to really tweak it (even if thats a preset that sets it up to do like version 0.5). But i still don't think this is the best route and i believe we can still aim for the one-sensitivity-slider-perfect-setup-for-all-solution.
Last edited by JamOrigin on Tue May 21, 2013 9:40 pm, edited 4 times in total.
JamOrigin.com

Like us on Facebook.com/JamOrigin and follow us on Twitter @JamOrigin

Post

Version 0.7.2 is out.

Whats New:
* Reimplemented plugin scanning to be less error prone.
* Added Add File button to plugin browser to be able to add single plugin.
* Added Clear button to plugin browser to clear the list of plugins.
* Fixed a bug in tuner that made it show too sharp

Download:
http://jamorigin.com/midi-guitar/MIDI-G ... .2-Win.zip
http://jamorigin.com/midi-guitar/MIDI-G ... .2-Mac.zip
JamOrigin.com

Like us on Facebook.com/JamOrigin and follow us on Twitter @JamOrigin

Post

JamOrigin wrote: AdmiralQuality,

Tuner should be fixed now. Obviously bad recognition will cause the tuner to jump around. More on this below.
No reason to go that low in buffersize. Optimal latency you get currently at 128 @ 44.1Khz. If its too hard on the system then go with 256 for now.
This is my ASIO latency in my host (Reaper). Nothing to do directly with your software, and obviously I have every reason to want that latency as low as it can possibly go. If I could set it to 1, I would!

Incidentally, your software is not reporting any latency back to the host. (In VST it's done by calling the setInitialDelay() function in the constructor).

Another question I've had all along is whether your MIDI events are being sent with negative deltaFrames values, so that the host can move them back to their correct place in time (on playback, obviously not possible in real-time).


Multichannel comming. Most changes in 0.7 and also the cpu overhead you experience is actually related to our preparation for multichannel midi. Maybe we gave it too long a leash and so there are some optimizations left for us to improve on.
Like I said, it should just be a few lines of code to do it simply. Determining which string is which note is nice, but only really necessary when you start wanting to assign different sounds to different strings.

Am I correct that it's producing 7 notes at once now? Is this intentional? If so, maybe a slider somewhere to define the number of strings so the software doesn't have to (incorrectly) guess? If not, then you've got a bug as I definitely saw 7 keys down at once in Poly-Ana.

As mentioned general the tracking of 0.7 seems to be much more dependent on your guitar/setup or playing style whereas 0.6 was very robust for most. This isnt very satisfactory, but we got some invaluable knowledge in the process. We have a few more things to try out in the search for the "silver-bullet". If it won't success we will just bring in the sliders, compressors and preset you need to really tweak it (even if thats a preset that sets it up to do like version 0.5). But i still don't think this is the best route and i believe we can still aim for the one-sensitivity-slider-perfect-setup.
Again, I found 0.6 to be frustrating as well when compared to earlier versions. We have to go back to 0.5 to find the last time it was fun to use. These last two (major) updates have just been unplayable for me, I'm sorry to report.

Hope you can get it back on track!

Post

0.7.2 cleared up scanning issues for me. Thanks

Post

scook wrote:0.7.2 cleared up scanning issues for me. Thanks
same here.big improvement in tracking also.tnx alot jamorigin :)

Post

Yes, v0.7.2 did clear up the VST scanning issue, as well as the polyphonic tuner issue. I've compared the polyphonic tuner's result with a couple of my tuner VSTs and the result is very close to them. However, I find the current "range" of the polyphonic tuner display too wide to function efficiently as a guitar tuner. Right now it has a display range of a bit more than +/- one (1) tone. That's too large to make precise tuning adjustments. Would it be possible to add a switchable option that can select between say 3 level of precision: 1) +/- 1 tone (like currently) 2) +/- 1 semi-tone 3) +/- 50 cent? I believe most users would like such flexibility.

As far as improvements in tracking compared with the previous version is concerned; well, after the initial excitement and a lot of testing, it's hit/miss in my case. Latency has been greatly reduced (from ~25ms with v0.6.x to as low as 8ms with v0.7.x) but I'm also getting far more problems; un-triggered notes, wrong notes been triggered (often a full octave above the real note) and notes cut-off if I play some riff on other strings at the same time. I've tried all kinds of pre-processing in my DAW to try improving detection (compressor, high EQ boost, etc) and although some of these improved detection of some notes, it worsens other notes. Overall, I could not achieve consistent and reliable tracking. From other people's feedback, I read that the model of the guitar and its pickup can have a big effect on trackabilty. Maybe it's because I have an Epiphone SG (STD pickups) that I'm having these problems? In any case. I hope MIDI Guitar will improve up to the point that it produces reliable results with any guitar.

Chuck

Post Reply

Return to “Instruments”