But in the JACK Audio Connection Kit I can select one of three options:
MIDI Driver none | raw | seq.

The MIDI Driver seq is the one that Pianoteq chooses automatically, once a Midi input is selected in Pianoteq (for example the EdiRol UM-1SX):

$ lsof | grep seq
Pianoteq  1656                    zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
Pianoteq  1656 1659               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
Pianoteq  1656 1660               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
Pianoteq  1656 1661               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
Pianoteq  1656 1664               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
Pianoteq  1656 2378               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
vpuptthd  1656 2379               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
tith      1656 2380               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
JUCE\x20T 1656 2381               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq
JUCE\x20M 1656 2443               zi    8u      CHR              116,1      0t0       9754 /dev/snd/seq

So I made two new tests with the Edirol - one with "raw" as before and one with "seq" - and tried to visualize the results in a spreadsheet:

(original data at the end of this post)

The midi latency with the driver seq is higher and smeared over a wider range.
This "smearing" around an average latency is usually called MIDI Jitter.

Although the difference between both MIDI drivers seems to be small, it would be nice if Pianoteq would have an option to choose the driver (raw or seq) in the GUI.

And/or in the CLI similar to jackd with options -Xraw and -Xseq.

Happy New Year!

PS:
I found a workaround to use the MIDI Driver raw in Pianoteq:
- Deselected all midi inputs inside Pianoteq and stopped it.
- Then started jackd with Midi driver raw and Pianoteq.
- Connected my USB MIDI clients (Korg B2 and a continuous sustain-pedal) virtually with the writeable client Pianoteq in the Connect pane.
Alas I can't use this workaround, because it produces more load on my fanless Intel and I have some Xruns (jackd now has to handle MIDI additionally - before I used MIDI Driver "none")

Original data:

=================================================================
MIDI Driver seq in qjackctrl
/usr/bin/jackd -p128 -dalsa -dhw:USB -r44100 -p64 -n2 -Xseq -P -o2

$ jack_midi_latency_test
Waiting for connections ...
Waiting for test completion ...

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 5.98 ms (263.59 frames)
Lowest latency: 5.40 ms (238 frames)
Highest latency: 6.65 ms (293 frames)
Peak MIDI jitter: 1.25 ms (55 frames)
Average MIDI jitter: 0.24 ms (10.07 frames)

Jitter Plot:
0.0 - 0.1 ms: 222
0.1 - 0.2 ms: 227
0.2 - 0.3 ms: 187
0.3 - 0.4 ms: 219
0.4 - 0.5 ms: 129
0.5 - 0.6 ms: 38
0.6 - 0.7 ms: 2

Latency Plot:
5.3 - 5.4 ms: 1
5.4 - 5.5 ms: 30
5.5 - 5.6 ms: 87
5.6 - 5.7 ms: 90
5.7 - 5.8 ms: 101
5.8 - 5.9 ms: 113
5.9 - 6.0 ms: 112
6.0 - 6.1 ms: 119
6.1 - 6.2 ms: 114
6.2 - 6.3 ms: 84
6.3 - 6.4 ms: 116
6.4 - 6.5 ms: 43
6.5 - 6.6 ms: 12
6.6 - 6.7 ms: 2

Messages sent: 1024
Messages received: 1024
==================================================================

MIDI Driver raw
/usr/bin/jackd -p128 -dalsa -dhw:USB -r44100 -p64 -n2 -Xraw -P -o2

$ jack_midi_latency_test
Waiting for connections ...
Waiting for test completion ...

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 5.61 ms (247.27 frames)
Lowest latency: 5.42 ms (239 frames)
Highest latency: 5.92 ms (261 frames)
Peak MIDI jitter: 0.50 ms (22 frames)
Average MIDI jitter: 0.09 ms (3.42 frames)

Jitter Plot:
0.0 - 0.1 ms: 615
0.1 - 0.2 ms: 396
0.2 - 0.3 ms: 12
0.3 - 0.4 ms: 1

Latency Plot:
5.4 - 5.5 ms: 150
5.5 - 5.6 ms: 379
5.6 - 5.7 ms: 236
5.7 - 5.8 ms: 239
5.8 - 5.9 ms: 19
5.9 - 6.0 ms: 1

Messages sent: 1024
Messages received: 1024

Principle is again "roundtrip". A test MIDI signal is sent via USB, converted to DIN-MIDI-OUT, physically looped back with a MIDI cable to DIN-MIDI-IN and back via USB to the computer. The delay/latency is analyzed by ...

$ jack_midi_latency_test
Waiting for connections ...
Waiting for test completion ...

Reported out-port latency: 0.00-0.00 ms (0-0 frames)
Reported in-port latency: 0.00-0.00 ms (0-0 frames)
Average latency: 5.61 ms (247.39 frames)
Lowest latency: 5.46 ms (241 frames)
Highest latency: 5.83 ms (257 frames)
Peak MIDI jitter: 0.36 ms (16 frames)
Average MIDI jitter: 0.05 ms (1.50 frames)

Jitter Plot:
0.0 - 0.1 ms: 936
0.1 - 0.2 ms: 87
0.2 - 0.3 ms: 1

Latency Plot:
5.4 - 5.5 ms: 39
5.5 - 5.6 ms: 314
5.6 - 5.7 ms: 632
5.7 - 5.8 ms: 37
5.8 - 5.9 ms: 2

Messages sent: 1024
Messages received: 1024

The reported average MIDI roundtrip latency of the Edirol UM-1SX is 5.61 ms.

For comparison: The internal USB-to-MIDI interface of the Korg B2 keyboard seemed to have a similar MIDI roundtrip latency of 6 ms.

I just speak for my system tested. The midi roundtrip times were reproducible, I tried it several times and with several midi notes. The audio from the usbmidi loopback has always been 6 ms behind the audio of the DigitalPiano triggered from the internal keyboard.

The USB-connected notebook had realtime scheduling and a linux kernel with 1000 Hz system-timer. The only significant application running was jackd with connection window open to loopconnect the B2's MIDI-OUT and MIDI-IN.
So not a system I would call "busy".

The key press was eliminated in my experiment, because the timing starts from the moment a MIDI note-on event is triggered (at the end of the key movement, when the last key contact has been closed).

But of course it is not more than one real life snapshot of the second half of your topic: "vs USB midi built-in".

End-to-end audio recordings (fingernail to speaker) were made several times in other threads of this forum. If you are more interested in those you can use the search function with string "latency" and user "groovy" ;-)

cheers

RobertS wrote:

I will get a Korg D1 with a midi to USB cable so I will be able to compare it with other keyboards I have with the the midi interface built-in and will share here my subjective findings.

Regarding the missing una corda and sostenuto pedals: I found out today that it is easy to add more pedals directly to the set-up by using a device called Midi Expression from Audiofront. . . .

Yes, if you're using your DP as just a controller for a VST like Pianoteq, it really doesn't matter whether the pedals run through your DP or not.  I myself have a Roland FP-10 which has sort of a crippled half-pedal functionality: it outputs only three values for the pedal, 0, 127, plus another value I can't recall, which is around 64.  To get full spectrum half-pedaling I just run my pedal through my Yamaha Reface CP, which does have full half pedaling.  True, it's not the most elegant solution. But then again, I already had the Reface CP, so my fix for half pedaling comes at zero cost.

It was not so easy to get some real values for MIDI-latency. As suggested in my last post I tried to measure the time lag between a played piano tone on a Korg B2 and the same midi event send to a PC and loopbacked to the B2 (via its USB-MIDI-OUT/IN).

I recorded the two tones from the Line Out of the B2 with Audacity and used its spectrum analyzer. Test tone was the default piano played gently with the leftmost key on the 88-keyboard.

The recorded wav is a komplex superposition of both audio signals (the delayed midi event returns while the key is still down). But there is a function called "Enhanced Autocorrelation", that can find correlated structures in a signal. Because the first and the delayed signal are very similar this function found this strong correlation at a time period of 6 ms "0.006 sec".

My interpretation is: The MIDI event took 6 ms more to arrive at the DAC of the Korg than the MIDI event directly from the keyboard to the DAC.

A rough estimation: The 6 ms were IN and OUT, so just for OUT circa 3 ms.
Probably >3 ms for OUT, because keyboard internally to DAC takes some time too.

Another unknown time variable is the loop connection inside the PC (MIDI raw device in the kernel short-circuited by jackd (virtual cable between B2 MIDI-IN and B2 MIDI-OUT)). I assume this to be <<1 ms.

In one sentence: I assume that the USB-MIDI-OUT of a KORG B2 has a MIDI latency of >3 ms.
.

Thanks for the interesting experiment. This implies the time it takes from the moment the key is hit til the moment the DAC plays the sound a constant. Also the loopback time is considered to be constant. But they may not be constants, specially the loopback time where the computer expectedly won't be as busy in both moments so it may not take the same time to perform the same task.

The point was to discuss the latency from the moment a key is press til the moment Pianoteq produces the sound. In a real acoustic the "latency" would be the time taken from the moment the hammer hits the cords til the moment the cords start vibrating producing sound. So there will be a little latency.

In my opinion the best way to measure the real latency of a midi keyboard would be to take a mobile phone, place it near the keyboard and start recording audio. Then press hard a keyboard key, so it will record the sound from hitting the key and then it will record the sound from the speakers playing the Pianoteq note. Then from the audio editor calculate the time taken from the moment of the first noise hitting the key til the moment the first part of the sound shows up in the waveform. Then deducting from the total time the output latency of the audio driver as reported by the DAW, it should give the real incoming midi latency of the system.

I recorded the two tones from the Line Out of the B2 with Audacity and used its spectrum analyzer. Test tone was the default piano played gently with the leftmost key on the 88-keyboard.

The recorded wav is a komplex superposition of both audio signals (the delayed midi event returns while the key is still down). But there is a function called "Enhanced Autocorrelation", that can find correlated structures in a signal. Because the first and the delayed signal are very similar this function found this strong correlation at a time period of 6 ms "0.006 sec".

My interpretation is: The MIDI event took 6 ms more to arrive at the DAC of the Korg than the MIDI event directly from the keyboard to the DAC.

A rough estimation: The 6 ms were IN and OUT, so just for OUT circa 3 ms.
Probably >3 ms for OUT, because keyboard internally to DAC takes some time too.

Another unknown time variable is the loop connection inside the PC (MIDI raw device in the kernel short-circuited by jackd (virtual cable between B2 MIDI-IN and B2 MIDI-OUT)). I assume this to be <<1 ms.

In one sentence: I assume that the USB-MIDI-OUT of a KORG B2 has a MIDI latency of >3 ms.
.

If your "feeling" doesn't deceive you, this could mean that the internal keyboard MIDI-to-USB has a better MIDI latency and/or better MIDI jitter than the external OTraki MIDI-to-USB converter (given all your other devices in your chain, like USB-HUBs and MIDI-Mergers are 100% identical).

If you are having spare time in the next weeks, you could make a more objective experiment: The idea is to loopback the MIDI-out to the MIDI-in. When the keyboards tonegenerator is MIDI-Local-On you should hear two sounds - the native and the loopbacked a few milliseconds later.

If you record this tones with a waverecorder (like Audacity) you should see two wavs with a few milliseconds distance. Which will have the shorter soundgap in your soundsystem - the DIN-MIDI-loopback or the USB-MIDI-loopback?

Thanks for your little experiment.

Got my Yamaha MODX8 that has both standard midi outs and a Midi thru USB. Connected both types of midi connection and set my Merging interface to 16 samples latency and Logic Pro X to 32 samples latency. Since from the Control Panel of the MODX it is possible to specify which MIDI I/O is using I was playing Pianoteq while switching both types of connections in the Yamaha.

My feeling was to favor all the time the connection thru USB (Even though this USB was connected to a 10 ports USB3 hub to the Mac Pro). The Chinese midi interface was decent and playable but playing very fast triplets for instance with the USB connection I was feeling connected to the keyboard and the sound, whereas with the midi to USB I was feeling sometimes unconnected. There is clearly a little more latency with the Midi to USB cable in this case. The question now is will a more expensive midi interface deliver better results? I don't expect so. It seems complicated to get a top midi keyboard for the Pianoteq playing;)

RobertS wrote:

I will get a Korg D1 with a midi to USB cable so I will be able to compare it with other keyboards I have with the the midi interface built-in and will share here my subjective findings.

Regarding the missing una corda and sostenuto pedals: I found out today that it is easy to add more pedals directly to the set-up by using a device called Midi Expression from Audiofront. You use it between an additional (continuous) pedal and the pc (so not the dp). Then map it to the number Pianoteq expects for (say) the una corda pedal. I haven't tried it yet, but they said it's pretty straightforward. But of course it would mean extra costs of an additional pedal + the Midi Expression. And with the additional connections going on it's a slightly less elegant set-up (if you care about that). There are single and multiple versions of Midi Expression, so you can even add more pedals.

Regarding the missing note-off signal: nothing can solve that except an urgent letter to Korg's CEO asking to issue a firmware update asap (should be only a few lines of code, because the piano is picking up the signal anyway) ;-)

Thanks! This sounds very interesting. Though I had not noticed that it only has one pedal socket for half pedaling! This along the fact that it lacks a built int USB port and it hasn't note off velocity it seems there are too many compromises for me to accept.

I will get a Korg D1 with a midi to USB cable so I will be able to compare it with other keyboards I have with the the midi interface built-in and will share here my subjective findings.

Regarding the missing una corda and sostenuto pedals: I found out today that it is easy to add more pedals directly to the set-up by using a device called Midi Expression from Audiofront. You use it between an additional (continuous) pedal and the pc (so not the dp). Then map it to the number Pianoteq expects for (say) the una corda pedal. I haven't tried it yet, but they said it's pretty straightforward. But of course it would mean extra costs of an additional pedal + the Midi Expression. And with the additional connections going on it's a slightly less elegant set-up (if you care about that). There are single and multiple versions of Midi Expression, so you can even add more pedals.

Regarding the missing note-off signal: nothing can solve that except an urgent letter to Korg's CEO asking to issue a firmware update asap (should be only a few lines of code, because the piano is picking up the signal anyway) ;-)

Cheers thiesdewaard!

You're welcome Qexl, glad to be of help!

All these issues disappeared or were vastly improved as I started using midi thru USB cables. But I always tried with vintage keyboards, never with a new one like the Korg D1 or the Kawai ES110 and a new midi interface. Guess it is a matter of trying!

I will get a Korg D1 with a midi to USB cable so I will be able to compare it with other keyboards I have with the the midi interface built-in and will share here my subjective findings.

the initial question is hard to answer. First you have to find a person, who has a keyboard-controller with two MIDI-Interfaces: DIN-MIDI and USB-MIDI.
Then an external MIDI-to-USB adaptor is required.

You have to trust that person, that it is able to "hear" MIDI jitter and to "hear" MIDI latency.

Even if you trust for some unknown reason, we don't know anything about the quality of the external MIDI-to-USB adaptor and quality of the internal MIDI-to-USB implementation in the keyboard. Take another adaptor and another keyboard and another person and the result may be completely different.

A measurement would be clearer. But who really can measure MIDI-jitter and -latency in situ?

Unless we have measurements, we can just try some generalization:

* It should not make a difference, if the MIDI-to-USB conversion is done inside or outside a keyboard. Probably the same (old & cheap) technology is used.

* Many MIDI-to-USB converters are using USB 1.x, which has a packet latency of 4 ms. DIN-MIDI has a resolution of ~1 ms, so USB conversion takes a few milliseconds of time more (minimum).

* MIDI-jitter might be higher when USB is included (randomly a MIDI-event reaches the current USB-packet or the 4 ms delayed next).