# Topic: velocity resolution

1. ## velocity resolution

so...

this is the third in a series of posts, it is related to the others but with its own focus, so I assume a separate thread is justified.

My question: how many different velocity levels does a pianist need? There are 127. My feeling is that I wouldn't need that many. But if I don't need them, this doesn't mean another doesn't either...

127 is quite a lot. Have you tried hitting a key with exactly the same velocity over and over? If I do, there is quite some variability in the resulting velocity value. Would we loose a lot if we reduced the number of velocity values so that the attempt to hit it with equal velocity would reliably result in the same velocity value?

The underlying idea is: the time may come when fading between samples (or volume adjustments or whatsoever) is no longer needed because we have enough samples per note to cover the required velocity resolution. (Different samples for all 127 velocity levels is not realistic - regardless of what I've just learnt about the announced VSL).

-mat123

2. ## Re: velocity resolution

Originally Posted by mat123
My question: how many different velocity levels does a pianist need? There are 127. My feeling is that I wouldn't need that many. But if I don't need them, this doesn't mean another doesn't either...

127 is quite a lot. Have you tried hitting a key with exactly the same velocity over and over? If I do, there is quite some variability in the resulting velocity value. Would we loose a lot if we reduced the number of velocity values so that the attempt to hit it with equal velocity would reliably result in the same velocity value?

The underlying idea is: the time may come when fading between samples (or volume adjustments or whatsoever) is no longer needed because we have enough samples per note to cover the required velocity resolution. (Different samples for all 127 velocity levels is not realistic - regardless of what I've just learnt about the announced VSL).

-mat123
I don't know how many velocity levels a pianist needs, but you are quite correct that achieving the same velocity level repeatedly is extremely difficult (at least I can't do it). Perhaps we don't need these minute gradations, but it's better to err on the safe side when setting up a standard, and apparently having too many steps hasn't been a problem.

Keep in mind that the basics of MIDI is pretty well a universal standard (some manufacturers have expanded the sound set), and the concept of a standard is very useful. If some manufacturer decided to reduce the number of steps, then their equipment would be non-compatible with other equipment - this is what a standard is all about.

If samples DID have 127 velocity/loudness levels then midi WOULD require all 127 level because the velocity value generated by the keyboard is how the correct sample is selected. If the number was reduced, the system wouldn't function in this case.

One final comment - it seems you are focused on samples being the source of our MIDI sounds and to my knowledge no one has a sample set with 127 or 128 levels of anything, certainly not velocity.

However, samples are NOT the be all and end all of MIDI sound. In fact, I predict physical modeling will replace samples, and perhaps sooner than many would wish. Physmod produces all 128 levels of velocity - no switching or blending between samples is required. There are at least two physmod software programs on the market, and Roland just announced their V-Piano which is not ROM based, but generates the sound with mathematical algorithms.

Glenn

3. ## Re: velocity resolution

Originally Posted by Glenn NK
However, samples are NOT the be all and end all of MIDI sound. In fact, I predict physical modeling will replace samples, and perhaps sooner than many would wish.
Glenn
Although this is not a modelling versus sampling thread, I can't resist to comment...

I am not afraid of the future, it will bring us even better sounds, no matter on which side you're on; but I bet that the near future will not decide between samplers and modellers. To my eyes, the future of modellers is much more uncertain than that of samplers. Modellers will improve and approximate a certain level of realism - and it's hard to tell how good this level is. On the other side, memory requirements of samplers will be a non issue in the not so far future, so there is no reason not to assume samplers will get to 100% realism...

cheers,
mat123

4. ## Re: velocity resolution

The discussion about the number of velocity boils down to this. More velocity resolution, means more accuracy. With 127, the accuracy of the mechanical parts and detection methods can either be less than adequate (detects less than 127), or at most, 127. I think it's quite true that you don't need all 127. But, unfortunately, many keyboards can't output full range, or the outputs seem truncated either because of manufacturing inconsistency or maybe even a defect.

However, if the manufacturing process was optimized for 16,000 velocity levels like VAX77, then the precision of the entire process would make minimize inconsistencies by a great deal. It's sort of like drive testing a family van on a speedway. If it's safe on a speedway, it's definitely safe for normal driving.

The other problem is that because a keyboard velocity range is quite limited, 127 can almost be too dense. By velocity range, I mean the force it takes to output a velocity = 1, to the force it requires to reach velocity = 127. Usually you can max out quite easily. So, subdividing that limited range will result in each of the 127 velocity values being very close to each other, effectively reducing resolution. (like, if 126 so close to 127, that the distinction isn't necessary)

What this problem relates to is that it seems like most sample libs are much to dynamic compared to the velocity dynamic range of the keyboard. If the max velocity of the sampled piano, is higher than the max velocity attainable by the keyboard it is being played back on, then the velocity relationship is broken. Meaning, since the sample lib is trying to fit in more dynamic range than the keyboard's velocity allows, the result will be a playing experience where the sample responds to keyboard velocity in an exaggerated way.

The problem is a tough one, but it seems to me it's an overlooked one because everone's out to create the most dynamic lib with the most layers. I believe that velocity relationship between the samples and the keyboard is the most important thing to a lib feeling good to play. Achieving maximum dynamic range in a lib is mostly a production concern since the result is of main concern, and not necessarily if it feels good to play or not. (ex: film scoring)

Side note on physmod.. I think physmod pianos have a good chance.. It's hard to say because physmod in general has a very bad track record. There's bowed string, plucked, brass, reed, flutes... most of these haven't really replaced their sampled counterparts yet, even after being available to the masses for nearly 15 years. Most of them sound kind of strange, and are a pain to use. But piano is different because the interface is right.

5. ## Re: velocity resolution

Check out Roland's newly announced V-Piano - it's the first full package that doesn't use samples. It models the sounds in real time.

The advantage of modeling is that the sound can be easily changed to achieve an infinitely variable range of different instruments. With samples, to get a "different" piano, you require another sample set.

I am able to accurately mimic harpsichords, clavichords, electronic keyboards, and even an acoustic bass in addition to more than a dozen different presets of acoustic grands. The total installed capacity on my HDD is just under 30 MB - for the traveling/gigging musician, this is a huge advantage.

6. ## Re: velocity resolution

Hello All,

It has been a great number of years since I heard about what I am about to write; the exact source of my statements escapes me, but this was with regard to real pianos. It was made long before the introduction of MIDI and sampled/modeled pianos were invented.

These figures are quoted for general information to supplement the content of this thread. They are offered here to serve as a rough guideline to discuss the topic of velocity resolution.

An amateur pianist with less than a few years of lessons might have about half a dozen somewhat reproducible velocity levels on a decent grand piano, and perhaps less on an upright piano.

Omitting upright pianos for the moment, a competent pianist who has played for a number of years may have perhaps up to 10 or 15 reproducible velocity levels on a decent piano.

Aspiring concert pianists trained in accredited university programs have control over between 15 and about 25 shades of nuance on a nicely maintained and regulated studio grand piano 7'+ in length.

World class concert pianists on professionally maintained and regulated regulated world class concert grands may have between 20 and 50 reproducible shades of nuance, including their use of expert pedaling techniques.

Cheers,

Joe -- jcfelice88keys

7. ## Re: velocity resolution

Just thinking loud here (and without any "scientific evidence");
In my view the greatest limitation to many sampled pianos today, (besides the mysterious "ploinky" sound that still haunts me) is the lack of velocity-levels. Even when the sound of the samples in itself can be quite convincing, I often notice abrupt changes in timbre on even the largest sampled pianos. Sometimes I need one particular note to be slightly louder and I raise the velocity-level by as little as 1 or 2 values in the sequencer. But that could mean jumping from one velocity-level to another and the timbre would be completely different. Also, a change from "pedal up" to "pedal down" could produce very exaggerated change in the sound.

I would say that for instance 10-12 velocity layers mostly are inadequate for reproducing anything totally convincing. Not that many pianists are capable of reproduce greater nuances intentionally, but even if you are intentionally incapable of reproducing a tone in the greatest detail, you will almost certainly hit the keys in a slightly different way every time you play and that way creating tiny nuances and small, unintentional differences in timbre. That is in my view one of the things that makes music sound "alive" and still separates a real piano from a sampled piano. The skill of the player is not that important in this respect. A bad sampled piano will sound bad regardless of the performance because it's not a real instrument. As long as the human ear is capable of hearing the difference between more than 12 variations of a sound, a sampled piano will certainly sound a bit fake.

8. ## Re: velocity resolution

Halvor:

These are some interesting thoughts you've presented - I haven't enough knowledge on the matter to support or refute them but they may explain why the demos on sample websites never sound convincing. Or why some notes are inexplicably too loud or too soft.

I listened to a rendering of Clair de Lune posted on a website (a Bosendorfer 290, but I won't identify the brand) and the upper notes of a passage of triplets near the beginning hardly sound at all. There are other similar instances of this in many sample demos available on the internet. It's particularly noticeable with classical music - perhaps because we've heard these compositions so often, and know what they should sound like. But when a note is too loud or too soft, it's quite obvious.

Glenn

9. ## Re: velocity resolution

glen, ya, I was speaking directly of v-piano. and the up coming pianoteq, and laso the pianoid dsp box. I do hope pianoid makes it to production.

Anyway, the discussion is going slightly off track.. but for demos, many companies post tweaked MIDI. Rarely do they put up tracks that are recorded with the player playing the actual sample lib.. so it really doesn't showcase how the lib would sound when played. It's a good demo if you use it for production, or if your hobby is to tweak MIDI. I know there are tons of people who spend hours and hours tweaking MIDI to sound good on a specific lib, but not necessarily play piano too well. Not really saying that's good or bad, but just saying it's a very popular hobby. To those people, it's like a customizable music box.

10. ## Re: velocity resolution

I had never heard of PianoID, and so checked out their website. My first thought was that this is a joke. Doesn't even sound like a cheap toy piano. If this is serious, then they are releasing info / demos WAY too soon. Why would they do that? I'm bewildered.

Right now I'm enjoying the Erard add on to Pianoteq 2.3. I'm looking forward to checking out the V-Piano. PianoID is not on the list - unless I'm missing something, here?

Lawrence

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•