Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Internal sounds vs. MIDI responsiveness

+1. Seeing that the dynamic range of an AP between each of 127 steps is already beyond the detectable range of human hearing pretty much ends the discussion for me. There are way bigger fish to fry in DP technology!

prout: I think you nailed it.

This topic has been discussed here quite a number of times in years past.

I have long understood that 1 dB is the smallest change in loudness that the human ear can detect ... and that only under the best conditions. Someone here countered that some people can distinguish .5 dB.

So yes, prout-san ... 127 levels is enough for .5 dB increments ... which is as fine as the finest ear can discern.
(I think you could cut that in half to 64 levels and VERY few would notice. Cut it in half yet again and few would notice!)

But increase it to 4096? That's just a techno-copout. Even the "ours-is-bigger" marketeers who trot out the polyphony numbers, and who declare that their keyboards rival the finest concert grands have not yet made touting mention of the virtues of MIDI hi-res. So I'm inclined to ignore this "feature".

JoeT points out that in some cases a production environment might need the hi-res to tweak things up for special purposes.

But I approach the piano as a performer, not as a technician. I don't have special purposes. I wonder how many people do? And are they in the piano market? Or are they in the production equipment market?

My personal experience is that when I had a PX-5S and connected it to Pianoteq using Hi-Res MIDI, I was amazed at how in-control I felt. It made playing Pianoteq a very pleasurable experience - so much so that I used to gig with that combination rather than use the perfectly adequate internal sounds. I suspect it had little to do with sound, and being able to detect minute differences in dBs (particularly on a gig), but everything to do with feel and feedback.

Of course, it was a subjective experience, but all I can say is that there was something different enough with that combination to draw me in more. Since buying a Roland FP-30, I no longer play Pianoteq live, as that sense of control is now missing for me.

I can draw no hard and fast conclusions about the value of Hi-Res MIDI, but personal experience would lead me to suspect that something in the finger-sound engine connection is improved with the increased resolution. Other factors may play an equal part, but I wouldn't write off Hi-Res MIDI as a gimmick.

vox: How does this experience differ from the situation where you feed low-res MIDI to Pianoteq?

I wish I still had the PX so I could test that scenario. Unfortunately, the FP-30 I have only outputs standard MIDI (as far as I'm aware), and I find it only average as a controller - not good enough to make me want to hook it up to Pianoteq in a live situation.

Maybe the PX-5S is just a very good controller board and would be equally good in low-res. It would be great if someone who plays subtly could check that out. All I remember is that subliminal feeling that I could express myself better with that combination.

Can you hear the difference in 0.5 dB?

http://www.audiocheck.net/blindtests_level.php?lvl=0.5

After some warming up, I absolutely can (>99% "confidence" according to the above site). To me, this means finer steps than 0.5 dB is generally important for music.

CDs use 16 bits (65,536 values) of resolution. I'm not totally sure, but I think this implies that many more than 128 volume levels for most sounds are accessible to CDs. This suggests that more volumes are well acknowledged to be desirable for music in general.

But pianos are not general instruments. Since they are percussive, it is less common than some other instruments to be need to hear many distinct volumes in a small volume-range.

For an electronic organ, though, you may want to smoothly change the volume output (say with a mod wheel) while holding some notes. Then very fine volume increments are significant.

If I played a descending five note scale multiple times with a decreasing volume and making the last note +/-0.5dB randomly, I seriously doubt that you could hear the change.

The test you referenced is just like the passive-acoustic coding tests done decades to establish the level of compression one could tolerate in the digital stream. In standard A/B note test, people could easily detect minute differences. In music, the threshold level for detection rose immensely. If this were not the case, MP3 would never have made it to the market. We hear what we want to hear.

With regard to bit depth, the depth simply provides a theoretical noise floor which allows the signal to noise ratio to be larger for a greater bit depth.

16 bits is perfectly adequate to reconstruct an analogue waveform and has a nice ~96dB dynamic range though even a symphony orchestra needs only about 65dB range and a rock band about 3dB range. It has little to do with resolution when dealing in acoustics.

I think you should not test weather you can hear a 0.5 dB difference. A better way would be to play a piece of music 2 times : unaltered and with notes randomly changed (+1). If you don't discover the change, the piece is well enough rendered, and the need to have a HiRes MIDI would be proven.

An excellent idea.

The question to me is less about whether a listener can tell the difference, but rather whether the player can tell the difference, as he's playing (i.e. one system seems to better track his intent, or give him a sense of greater control). A listener can't know whether a given note should be louder, softer, or the same as the previous note, only the player knows. And I think the player is much more likely to sense a subtle difference as he's playing than the listener is in listening to a playback.

I think there are some passages of music where particular notes are repeated at nearly the same volumes. Like in minimalist music, however important fine volume distinctions there are. Personally, I think I could live without finer resolution than 0.5dB, especially since I don't usually listen to music where such situations often occur.

And what passes for general-market acceptance surely should not be given too much respect in an enthusiast context.

But I find the 128 midi volumes to be sufficient for my DP, as well as the far fewer volume samples per note in the Garritan CFX.

I play principally on an very fine acoustic grand. From ppp to fff I sense maybe 5 or 6 velocities that I consciously use. I am sure I use many more than that in a subtle way, but I highly doubt it would be more than maybe twenty levels.

It would be an interesting experiment, to map the 127 values down to 20, and see if you notice any diminishing in the subjective responsiveness.

Just a guess, but I think we might notice the change. There are good arguments being put forth here that there are subtleties when playing that go beyond our obvious notice. I can see both sides of the argument but lack the knowledge to test them out.

Haha, this is a really interesting idea... I'll write a Reaper script to do that at some point this week.

What bothers me about this whole hi-res MIDI thing ...

Why won't they just make pianos that sound more like pianos, instead of working on useless crap like hi-res MIDI?

How about putting in better speakers?
How about putting in better samples?
How about doing SOMETHING WORTHWHILE?

Maybe because one is easy and cheap and the other is hard and expensive?

In my experience, 128 steps for MIDI is sometimes not enough - although I think 1000+ steps is overkill. I've done a lot of mix downs on DAWs and workstations for large audio projects and plugins and the digital mixing desks will usually be limited to 128 steps. There have been a lot of times for me when I've had to blend over a dozen instruments and found that having, say a trumpet at 78 was too loud and having it at 77 was too soft. I needed to get it in between those values to get it just where I wanted it. That usually resulted in me having to adjust a bunch of other parameters to achieve it in a different way - or in the case of an audio waveform, I would sometimes normalise the waveform. So I do support more resolution for MIDI - although I'm not sure how significant this would be for playing a digital piano.

In the quest to optimize, "the feeling of connectedness," we might try to identify significant factors and irrelevent factors.

I suppose we could start with a few macro buckets for VIs, such as:

1- Physical - realistic key action.

2- Time domain - low latency and low jitter.

3- Media - realistic sound production from samples and/or modeling.

4- Audio - realistic sound reproduction by amplifier and transducers.
_____

Assuming one has: a good digital piano action, optimized computer performance in the time domain, and a good headphone setup, I think the greatest area for potential improvement is "Media."

Interestingly, I tried the Roland V-Drums tonight with some cheap headphones. I think Roland absolutely killed acoustic drums here (but it is expensive and the cymbals are mediocre). The physical mesh surface is superb, latency unnoticable, and has realistic sounding media. Roland used "high resolution velocity" of 317 which might have made for superb dynamics. Comparing an acoustic snare to V-Drums, I "think" I can hear what is going on. Perhaps Roland make the sound so interesting by adding:

a. slight randomness in velocity
b. (lots of) round robin "samples"
c. no randomness in timing

A snare drum has got to be much easier to replicate than a grand piano is. Maybe these factors also are critical for digital piano reproduction. What if we apply them to Garritan CFX? For me, Garritan CFX loses a bit of realism when I practice a short section and the sound is identical over and over again. A few ideas that might work with a DAW:

a. Introduce slight randomness in velocity to each note.

b. Introduce say 5 round robin "samples". Not quite sure how to do this. Maybe we could:

*** run 5 instances of Garritan CFX, each with slightly different settings (eg pitch, mic gains, sympathetic resonance etc.) but that would require a monster computer.

*** run 2 instances of Garritan CFX with slightly different settings. Then have the DAW "randomly" mix the output.

*** run 1 instance of Garritan CFX and then have the DAW "randomly" add some minor effects. Maybe EQ would work better than say panning or reverb?




ENDNOTES
_____

1 - AvantGrand action replicates real piano action; expensive but does not provide same feedback as real piano does. Several of the actions by the big firms are quite good. Regardless, there are only a handful of action options.

2 - I measured excellent (true round-trip) latency from the internal sounds of a few digital keyboards; the result of dedicated electronics and good factory optimization. Getting similar latency levels from VIs (Pianoteq & Garritan CFX) required a lot of Windows optimization, a powerful laptop, and a good interface. I did not measure jitter but several users have measured good results with the RME ASIO drivers. VI users face endless hardware-software-optimization options.

4 - Reproducing audio at realistic volumes and dealing with haphazard room acoustics is challenging; headphones are a pretty good and easy solution. A few inexpensive headphones are quite popular on the pianoworld forums.

I don't know if a digital piano would benefit from 317 velocity steps. The snare was fantastic from ppp to fff and maybe more responsive than a real snare drum. The Roland manual shows one can reset the V-Drums to 127 steps (bank 88) and try it out but I don't know how to do that.
_____