I usually play with 192Khz and that's what gives me the lowest perceived latency (I try playing a note with my digital piano with volume on to compare better).

I'm new to DAWs, but I've noticed that when I use Pianoteq in REAPER, the latency is gone. It's like using the onboard piano sounds in my DP. Not sure why this is though. The only thing I can tell is it uses 48Khz if I remember correctly. But that sample rate gives me a higher latency if I use Pianoteq standalone... Anyone knows why?

Block size,  buffer size and driver option in Reaper.  The numbers are probably set lower in Reaper.

You want to set this to the smallest you can get away with - then increase it slightly until you just think you might notice latency again, then back it down slightly until it's gone to find the sweet spot with a bit of headroom, and no noticeable latency.

Possibly the driver Reaper uses - check if Reaper isn't using a slightly different one to the slower Pianoteq standalone.

You should be able to get away with a smaller buffer margins in the standalone because you are only dealing with only 1 to 3  instrument tracks (1 instrument, or up to 3 in Standard and Pro layering).
Whereas when you work in the DAW slightly perceivable latency may have to be tolerated in order to cope with say 20 tracks or so, or maybe hundreds in your orchestral Magnum opus!

I'm new to DAWs, but I've noticed that when I use Pianoteq in REAPER, the latency is gone. It's like using the onboard piano sounds in my DP. Not sure why this is though. The only thing I can tell is it uses 48Khz if I remember correctly. But that sample rate gives me a higher latency if I use Pianoteq standalone... Anyone knows why?

The best MIDI Keyboard is the FLK MK23 Keyboard, he simulates on the keyboard mechanically where the action of a real Grand Piano triggers the hammer and uses high resolution velocity (4096 instead of 127) to capture finer details for PianoTeq to render the pitch/tone/dissonance/color, etc.

I went from a 128 velocity keyboard to his MK23 (with the high rez MIDI option) keyboard the PianoTeq sounded way more colorful and alive.

Thanks for your feedback about the FLK mk23.
It is very interesting to have here an owner of this “famous” (for educated people, though) keyboard.
As an owner, do you have any technical details about how it is supposed to simulate the way a grand piano triggers the hammer and particularly hit the string ? Is there any mention of an adaptive delay induced by the incoming velocity or real time speed ? This would be invaluable to know.

Also, some side questions please :
- Are you able to trigger consistently a similar velocity value when hitting many times the same note at the strength ?
- Are you able to repeat quickly one note without side effects like missed note-on or “hot” velocity value (ie. 127 suddenly) ?

Thank you !

I went from a 128 velocity keyboard to his MK23 (with the high rez MIDI option) keyboard the PianoTeq sounded way more colorful and alive.

Paulo164 wrote:

gabe wrote:

I would be very interested in seeing what you find. The way I did it is to use a second computer for measurement. I.e. use you normal computer for pianoteq and not for measuring. The measurement computer had a sound device where I recorded the analog output of the piano and at the same time either the analog sound output of the Pianoteq computer or a microphone for acoustic. If you have more than two inputs on your recording device you can do all three at once. Keep mic not too far from strings (a meter already adds 3 ms), in the middle. It helps a lot if the environment is silent (i.e. you only have your piano sound in the mic recording). It helps if you have both a spectrogram and wave form for measurement, but timing should be determined from the wave form. Low frequency noise can mess up the acoustic measurement, so you could filter that out, but make sure you filter all signals the same way. Good luck!

PS For those interested, there is an interesting paper on temporal dynamics of a grand piano here: https://www.researchgate.net/publicatio...no_actions

Hello,

Dully noted. However, I don't intend to measure the absolute latency of my setup. The purpose is to measure how the "acoustic" latency increases/decreases according to the incoming velocity. So I will setup on purpose a high latency on the digital part so that the 2 peaks ("acoustic" and "digital" ones) are well separated on the timeline, making reading the measurement easier.

In a second time, I will do a measurement with the minimum possible digital latency to see if I can get as close as possible to the latency of my acoustic piano.

Just to let you know I have done some measurements but it appears there is 2 peaks due to 2 phenomenons when pressing a piano key : first the hammer noise (quite tiny), then the string noise (much louder).
I think that the acoustic piano "latency" should be measured as soon as the hammer noise appears. So I must do again my measurements with this in mind, otherwise it's pointless.
I will keep you posted.

Sorry to come back so late to this thread !!

Here are my conclusions :
Depending on the speed you depress a key, there is an offset between the time the hammer in an acoustic piano hits the string and a MIDI controller triggers the sensor to produce the Note-On event.

Let’s assume the string and the sensor are triggered at the same time when you play forte. Then you will have a delay up to somewhere 5 to 10 ms when you play softly for the digital piano. Said differently, the digital piano is late compared to the acoustic one when playing softly.
This delay increases from 0 to 10 ms when velocity decreases from 127 to 1.

I explain this phenomenon by the fact the digital piano triggers the note-on always at a fixed distance of the key to the keybed, whereas on the acoustic piano there is a coordinate/speed equation much more subtle occurring.

As a consequence, a perfect digital piano should trigger very early its note-on events and apply a variable delay depending on the velocity.
By doing so, the main issue is you will loose precision because triggering early the sensor implies you measure the speed on a shorter distance.
So it is a trade-off to find here between accuracy of the measured velocity and accuracy of the timing.

Please note that I purposefully ignore the system latency in this reasoning. This is out of scope.

Thanks.

gabe wrote:

Paulo164 wrote:

I also have a kind of transacoutic piano. So will be able to measure the timing between the "acoustic" peak and "digital" peak, according to different velocities.

I would be very interested in seeing what you find. The way I did it is to use a second computer for measurement. I.e. use you normal computer for pianoteq and not for measuring. The measurement computer had a sound device where I recorded the analog output of the piano and at the same time either the analog sound output of the Pianoteq computer or a microphone for acoustic. If you have more than two inputs on your recording device you can do all three at once. Keep mic not too far from strings (a meter already adds 3 ms), in the middle. It helps a lot if the environment is silent (i.e. you only have your piano sound in the mic recording). It helps if you have both a spectrogram and wave form for measurement, but timing should be determined from the wave form. Low frequency noise can mess up the acoustic measurement, so you could filter that out, but make sure you filter all signals the same way. Good luck!

PS For those interested, there is an interesting paper on temporal dynamics of a grand piano here: https://www.researchgate.net/publicatio...no_actions

Hello,

Dully noted. However, I don't intend to measure the absolute latency of my setup. The purpose is to measure how the "acoustic" latency increases/decreases according to the incoming velocity. So I will setup on purpose a high latency on the digital part so that the 2 peaks ("acoustic" and "digital" ones) are well separated on the timeline, making reading the measurement easier.

In a second time, I will do a measurement with the minimum possible digital latency to see if I can get as close as possible to the latency of my acoustic piano.

Just to let you know I have done some measurements but it appears there is 2 peaks due to 2 phenomenons when pressing a piano key : first the hammer noise (quite tiny), then the string noise (much louder).
I think that the acoustic piano "latency" should be measured as soon as the hammer noise appears. So I must do again my measurements with this in mind, otherwise it's pointless.
I will keep you posted.

Paulo164 wrote:

I also have a kind of transacoutic piano. So will be able to measure the timing between the "acoustic" peak and "digital" peak, according to different velocities.

I would be very interested in seeing what you find. The way I did it is to use a second computer for measurement. I.e. use you normal computer for pianoteq and not for measuring. The measurement computer had a sound device where I recorded the analog output of the piano and at the same time either the analog sound output of the Pianoteq computer or a microphone for acoustic. If you have more than two inputs on your recording device you can do all three at once. Keep mic not too far from strings (a meter already adds 3 ms), in the middle. It helps a lot if the environment is silent (i.e. you only have your piano sound in the mic recording). It helps if you have both a spectrogram and wave form for measurement, but timing should be determined from the wave form. Low frequency noise can mess up the acoustic measurement, so you could filter that out, but make sure you filter all signals the same way. Good luck!

PS For those interested, there is an interesting paper on temporal dynamics of a grand piano here: https://www.researchgate.net/publicatio...no_actions

Hello,

Dully noted. However, I don't intend to measure the absolute latency of my setup. The purpose is to measure how the "acoustic" latency increases/decreases according to the incoming velocity. So I will setup on purpose a high latency on the digital part so that the 2 peaks ("acoustic" and "digital" ones) are well separated on the timeline, making reading the measurement easier.

In a second time, I will do a measurement with the minimum possible digital latency to see if I can get as close as possible to the latency of my acoustic piano.

I also have a kind of transacoutic piano. So will be able to measure the timing between the "acoustic" peak and "digital" peak, according to different velocities.

I would be very interested in seeing what you find. The way I did it is to use a second computer for measurement. I.e. use you normal computer for pianoteq and not for measuring. The measurement computer had a sound device where I recorded the analog output of the piano and at the same time either the analog sound output of the Pianoteq computer or a microphone for acoustic. If you have more than two inputs on your recording device you can do all three at once. Keep mic not too far from strings (a meter already adds 3 ms), in the middle. It helps a lot if the environment is silent (i.e. you only have your piano sound in the mic recording). It helps if you have both a spectrogram and wave form for measurement, but timing should be determined from the wave form. Low frequency noise can mess up the acoustic measurement, so you could filter that out, but make sure you filter all signals the same way. Good luck!

PS For those interested, there is an interesting paper on temporal dynamics of a grand piano here: https://www.researchgate.net/publicatio...no_actions

Paulo164 wrote:

gabe wrote:

- The internal DP engine is on average 10 ms faster than the acoustic sound (yes you read that right :-) ). This depends on key velocity though.  For fast velocities (loud notes) the DP engine it is ~5 ms faster, for low velocities (ppp) it is ~25 ms.

Interesting... This goes the opposite way of what the article pointed by @YvesTh states.

I read that too...
The ideal would be to make a slow motion video of a complete piano mechanism in operation at different velocities (pp,p,mf,f,ff) in order to verify the chronology of events (key stroke, key bottom contact, hammer stroke, string contact etc...)
And it is possible that upright  and grand piano are différents...

It would be great if Modartt could provide us with such a video.

I also have a kind of transacoutic piano. So will be able to measure the timing between the "acoustic" peak and "digital" peak, according to different velocities.

Paulo164 wrote:

gabe wrote:

- The internal DP engine is on average 10 ms faster than the acoustic sound (yes you read that right :-) ). This depends on key velocity though.  For fast velocities (loud notes) the DP engine it is ~5 ms faster, for low velocities (ppp) it is ~25 ms.

Interesting... This goes the opposite way of what the article pointed by @YvesTh states.

I read that too...
The ideal would be to make a slow motion video of a complete piano mechanism in operation at different velocities (pp,p,mf,f,ff) in order to verify the chronology of events (key stroke, key bottom contact, hammer stroke, string contact etc...)
And it is possible that upright  and grand piano are différents...

It would be great if Modartt could provide us with such a video.

It may be particular to the sensor system of the ATX3 system (and Novus NV-5, NV-10). It measures the hammer travel, not the key. I think between two points. The last point is going to be before let-off, so the velocity measurement is complete before the hammer hits the string (or stop rail).

gabe wrote:

- The internal DP engine is on average 10 ms faster than the acoustic sound (yes you read that right :-) ). This depends on key velocity though.  For fast velocities (loud notes) the DP engine it is ~5 ms faster, for low velocities (ppp) it is ~25 ms.

Interesting... This goes the opposite way of what the article pointed by @YvesTh states.

I read that too...
The ideal would be to make a slow motion video of a complete piano mechanism in operation at different velocities (pp,p,mf,f,ff) in order to verify the chronology of events (key stroke, key bottom contact, hammer stroke, string contact etc...)
And it is possible that upright  and grand piano are différents...

It would be great if Modartt could provide us with such a video.

- The internal DP engine is on average 10 ms faster than the acoustic sound (yes you read that right :-) ). This depends on key velocity though.  For fast velocities (loud notes) the DP engine it is ~5 ms faster, for low velocities (ppp) it is ~25 ms.

Interesting... This goes the opposite way of what the article pointed by @YvesTh states.

- The internal DP engine is on average 10 ms faster than the acoustic sound (yes you read that right :-) ). This depends on key velocity though.  For fast velocities (loud notes) the DP engine it is ~5 ms faster, for low velocities (ppp) it is ~25 ms.

- With a Pianoteq buffer setting of 128 (with Focusrite 2i2 sound device), latency of average would be comparable with the acoustic, and hence ~10 ms slower than the internal engine.

Conclusion: 128 buffer setting can be fine with dedicated sound device if you want to take the acoustic as a reference.