Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Latency in acoustic and digital instruments

Eight notes/sec is considerably faster than I can play except possibly for a trill. So it is quite interesting that you can hear and associate the finger to notes at that speed. I would have thought that the brain would quickly adapt to a delay of 50 ms, but the fact that you could notice it is a good data point. When I play short fast notes, I have no concept of whether the note is sounding as my finger goes down or as it goes up. So this may just show how different individual are and probably why threads like this usually produce a range of opinions.

I think a mismatch between the action and the sensory feedback by the consequences of that action is indeed the source of feeling 'disconnected'. The brain constantly monitors sensory input in relation to its own intended motor gestures, and produces an 'error signal' when sensory feedback does not match the prediction. This is fundamental to skill learning. The prediction depends on previous experiences.


But since the brain can learn the new (delayed) relation between motor gesture and sensory feedback, it may become better playable after a while, and become the 'new normal'.

IMO an underestimated factor in VST latency/disconnection/despair discussions is *random* variability in latency. This variation is unlearnable and will cause prediction errors all the time. By contrast, the latencies on an AP for different key velocities are variable but perhaps better predictable, and apparently sufficiently learnable because no one reports feeling disconnected to an AP.

I have seen (and provided) many latency measurements here on PW but these are always a single number like 10ms, which is either reported by the computer or measured as the difference between sound output in a handful examples. To get better insight into what's going on, it would be really nice if we had an idea of the variability in overall latency (i.e. action to sound output) on real-world systems.

One of the studies Del Vento lists finds that variation (jitter) as little as 3 ms around an average delay of 10 ms already causes subjects (even non-musicians!) to judge it as of 'lower quality'.

True, but how have these numbers been measured? Just what the computer reported, hence just the buffer latency, not the end to end? If so, they are of little interest.


I would love to be able to measure that. Unfortunately, as I have described, I have troubles measuring even a single latency number, let alone many latencies to estimate its jitter.

Please read what I've twice written above.I measure the TIME DIFFERENCE between the piano sound and the VST sound.

Sorry for missing that part of your message and thanks for repeating it. This is very interesting and consistent to my experience with the few tests I've run with a MacBookPro and a oldish but beefy Linux box -- without being able to measure the absolute latency per my discussion in my previous message. Setting Pianoteq buffer size for a reported latency of more than 5ms was definitely noticeable and more than 10ms was annoying.

How did you measure that difference? I suspect with something similar to what I described in my other message? Is there any way that you can turn down the volume of the piano sound to make it comparable to the key bottoming noise, and measure the latency of the internal piano sound, which should be pretty close to the difference between the key bottoming noise and the piano sound itself?

Do you have any way to do the measurement a few times to see if the jitter, as pianogabe suggests, that is "more annoying" than latency itself?

Thanks!

I am thinking out loud here, but previously I measured DP internal vs VST latency the way you also did this, by recording both simultaneously in audacity (or equivalent) and then manually look at the difference in 'sound start time'. I guess one could produce a script to create a midi file that plays the same note for a few hours, record both internal and VST output, and then use some simple threshold algorithm to find the note beginnings in the audio file. Since the note is always the same, and there is little noise involved (make sure output level is high), that might work sufficiently. If it doesn't one could use crosscorrelation or so. If you allow for more complex programming, you could use different notes and levels. Of course, because of different 'attacks' you can only measure variability for specific notes and levels.

This could also test the assumption that the internal DP engine does not have any appreciable latency variation.

Sounds like a nice programming project

That's what I did ... recording piano and VST in Audacity.

As for the programming/automation ... I'll leave that to others. I'm retired from tech.

Jesus Christ! Have we not measured enough already (pivots, keys, music rests, sample length/decay time, bench height/width, coffee table, piano apparel, etc.)?

I ask again, why do we have to measure the heck out of everything? Is there an underlying collective obsession that drives us to do this? Are we compensating for a collective want we do not -all- have? Is tiny, nano, small ever enough? Why? Why?

Why measure? Because my ears said "this stinks". So I measured. It stank.
I got better equipment. My ears said "this rocks". So I measured. It rocked.

That was years ago. Since then ... no more measuring.
But when someone asks, I respond.BTW ... Jesus measured. A lot! He was a carpenter.

The advantage of a measure is that it is objective. Let’s say you hesitate between a Steinberg UR22 and a Focusrite 2i2... latency measures can help you choose if you are very picky about latency.

If I told you my UR22 is Ok (for me), you won’t learn anything about if it is Ok for you.

And it is rare to find a music shop with both audio interfaces ready to be compared... sometimes, you have to trust some other opinions (it is ok), interpret some specs (well a 4ms latency should be ok), and perhaps make some mistakes ! (But I would prefer to make a $150 mistake about my audio interface than a $8000 mistake about my N1X... Surely the later was bern tested in a shop !).

I think I share your sentiment, and will shamefully go back to playing my piano in minute. But rationally I agree with Frédéric that this objectifies things and can be a great help, e.g. to see how to minimize latency if it appears to bother you. That and MacMacMac's argument that Jesus is on our side is also compelling.

Imagine a little boy hitting a makeshift ‘piano key’ and then never ever hearing a corresponding sound, a note, or even a hiss; well, imagine no more because ‘that little boy was me’!

Yes, it’s easy to take latency for granted when all you have to wait for is a few milliseconds to hear something.

Pete never had that luxury; growing up playing on a keyboard painted onto a wooden table. Yes, my dad was so poor that all he could afford for me was a can of white, a can of black, and an old crooked wooden table. That’s all that little boy had for a ‘piano’. So you see, that little boy played an ‘F’ but he never heard that note because latency was, by definition, infinite. The note never sounded, you guys, it never sounded.

So yes, I take a little offense when I hear some ‘round here complaining about a few milliseconds when that little boy had to wait till he grew up and bought a P-515 to ever hear that ‘F’ ring. Shame on you!

I can relate with you. I am an adult piano learner because of a similar situation: when I was kid I did not have a piano and my piano teacher told me to draw a keyboard on something like a roll of toilet paper which according to her would have been much better than the toy organ I had (that was basically a melodica with an electric pump). So I did not play neither the toy organ nor the toilet paper and I went to the lessons not prepared. Soon enough I gave up and here I am learning as an adult.

On the other hand, there are poor people today. Should we all sell our instruments to feed them? Or should we learn to play well to honor the Lord as J.S. Bach did? Since you mentioned Jesus in a previous message, this is a truly serious moral question. But it is OT in this thread, as we always end up doing...

Assuming we want to learn to play well, the question remains on what is the best instrument we can afford? Measurements can tell at least part of the story.

Pete ... All this talk reminds me of The Three Yorkshiremen.

For this particular question, you can do the following:

- Adjust DP volume to become the loudest sound source
- Record DP sound plus action noise plus ambient noise with a microphone (as you did)
- Simultanously record line-out from the DP (void of action/ambient sound)
- Normalize and correlate the two signals to find the best match. This indicates the time-relationship between the two recordings.
- Find the first indication of generated sound in the line-out recording. This indicates when the DP "knew" that a key has been pressed.
- Find the first indication of action noise in the microphone recording. It won't be buried below the piano sound because it's earlier.
- Possibly compensate for different sound travel time from action/piano to your microphone.

To be more exact about the action noise detection, you can create multiple recordings of the action noise and "average" them to create a synthetic ideal sound patch (once). Position the microphone to reduce noise that you repeatedly create with your fingers/body/bench. Or use different playing techniques for different samples so that unavoidable noise will average-away in the final result. Correlate this ideal reference sound patch with your recordings to extract the action timing.

You will probably get problems due to overlap between action and piano sound. You can trim the action reference patch to be shorter then the expected latency. Then there is no overlap. Likewise you can trim (all) line-out inputs by an arbitrary amount (from the easy to identify sound) and correlate only the remainder with the microphone input. For this, consider the unique sample length of your DP and when it starts looping the same waveform over and over (expect 1-4 seconds, or consult the corresponding megathread here on PW). Also, as recommended at the beginning, adjusting DP volume to be high probably makes this unnecessary.

To measure acoustic instruments, you can use a visual approach. Open the instrument and position a camera to capture both finger/key as well as hammer/string in a single frame. Capture a clock for a minute to make sure your nominal framerate is acceptable. Wobble the camera during a test capture to make sure it produces no tearing. If there is, the lines that compose the image are not being captured at the same instant of time. In this case, rotate the camera until all relevant components are aligned horizontally in the output (ideally all are visible in one single line). Record your data, then analyze the frames to count the time from touch to string.

Hammer travel time depends on velocity. You can use a player piano (pneumatic self-playing pianos like seen in western movies) to generate consistent velocities. Alternatively you can use a piano with silent system and MIDI out to read the actual velocity and reject recordings that fall outside of your target range. Lacking all, you may need to construct a device to get repeatable strikes (youtube has examples).

DPs also adapt to velocity, although probably less pronounced. The audio "peaks" may be less aggressive, or artificial delay may be introduced. For fair comparisons you should also control velocity on DPs.

I know this response is very pedantic, but you asked for it didn't you?

The closest I know is MacMacMac's experiment where he recorded both Piano output and VST output in audacity and then compared the waveforms.

One idea if you have a high FPS video camera (such as some of the Go Pro that can do 240fps - so one frame almost every 4ms) with a good mic. If you record a video and audio and then have a software that can go through video frame by frame and also show the audio waveform, you may be able to calculate the time from beginning of the key going down to when the sound is heard (which should be a very distinct peak in the audio waveform). Perhaps a slo-mo mode in a high end smartphone can be used for the same purpose. I have a iPhone 12 Pro which allegedly can do "Slo-mo video support for 1080p at 120 fps or 240 fps".

Osho

Your reasoning is too quick. You're taking incorrect shortcuts