Peter - I'm fairly sure you're wrong on this. Let's forget about quality for a minute and just on what MIDI Velocity value increase means for, say, a sampled piano - the VSL one for example.peter_schwartz wrote:Sorry Kevin, but no, you can't. First of all...Kevin Nolan wrote:Actually I think you can because the volume level increases with velocity relate directly to loudness of a piano.peter_schwartz wrote:No, you can't equate MIDI velocity steps to dB.
1) which piano are we talking about? A studio upright? A 6' grand? 7' grand? 9' grand? Which maker? My point here is that there is no standard for piano by which gradations of loudness are measured. Similarly, there are no standards for calibrating loudness on dB meters for which a piano of any kind is used.
2) what you and I perceive as volume increases in acoustic instruments are not just volume increases. They're increases in volume AND harmonic content (brightness). The ear, being more sensitive to higher frequencies than lower frequencies, will translate (in your mind) a brighter sound (with increased high frequency content) as "a louder sound". So if we played a low C on (any) piano at mF and then again at forte, you and I might agree that the forte note sounds 2x louder. But on a meter it wouldn't show that. It might be 1.5 times louder. That's because the extra brightness produced by striking the note harder gives us the perception that the sound has increased in volume by 2x. But if taking about dB's, it hasn't increased by 2x.
3) The resolution of MIDI velocity itself has zero -- I repeat, zero -- bearing on the dynamic range of any sound. If you're interested to know why, post back and I'll explain.
I'm sure that when VSL sampled this piano, they sampled each of the 128 velocity levels at a different loudness across the piano's loudness range. In other words, greater actual keyboard stroke velocity maps to the piano loudness. So they are mapping 128 MIDI Velocity values to what the piano has been designed to do louder over 128 steps from (probably) pppp to fff (or ffff) ie - sound louder. So although MIDI velocities are a linear scale, they map to the logarithmic scale of loudness -including the harmonic increases, ( as you say - part of what it means to be louder). It has nothing to do with the kind of piano or about you and me - loudness is not a percpetive scale per say - it' a straight forward difference between sound pressure levels and is independent of the listener. You are correct in indicating a perceptive element to this however, in that we can perceive something to be loud even if its volume is low (as in replayed at low volume or if heard at great distance - attributes such as envelope and harmonic brightness can tell us that it is being replayed at a loud level for that instrument even if we do not hear it loud there and then). But this is not directly related to our issue here - because even if you play back the VSL sampled piano at low volume on headphones, the 128 MIDI Velocity levels still match to the loudness levels originally sampled, giving you the correct sense that the loudness dynamics of the piano have been correctly captured.
In essence, it's the worldwide accepted standard of what a piano is and what a quiet and loud piano means that they have mapped that to MIDI velocity levels. They certainly have not mapped the 128 MIDI levels to linear sound pressure increases, so i thas to be a map of loudness.
This is related to why, for example, we can perceive an orchestral mock up as being validly loud if it uses samples of instruments sampled loud (and when replayed even at low volume) and why a sample orchestral mockup can sound wrong even at high volume if it's meant to sound loud but uses samples of instruments that were originally sampled quiet. This is particularly important to brass samples - you cannot achieve convincing brass mockup performances by playing soft samples loud - you need to sample the instrument at many loudness levels and attach each one to a velocity or CC controller level. Hans Zimmer did this famously for his brass in his original LSO sample set where the French Horns 'open up' louder as you move the modulation wheel - where different sampels at different loudnesses are invoked (I know this because I got temporary use of a version of his original orchestra sample set once !! )
It's also, by the way, why the SY77 and SY99 RCM synthesis is so incredibly acoustically convincing - by convolving FM with samples, you can use the harmonics in the FM sound to harmonically brighten a PCM sample as you hit higher velocities or change CC values, making it sound louder in a very acoustic way - it works fabulously on brass and woodwind instruments (the SY99 French Horn is amazing in this way). Here, albeit in the pure synthesis domain, loudness as we understand its associated characteristics (such as harmonic brightening) can be implemented through RCM synthesis and with loudness mapped to MIDI Velocity values! I haven't gotten into it enough yet but there may be a chance that MOD-7 can provide for this too.
In any case this is not the central debate - rather I still believe the point is valid that thousands of gradation steps in velocity sampling and available MIDI velocity levels are needed to deliver superior nuanced pianos that will approach achieving 'acoustic tone' in playing. That nuance has to be captured digitally because we notice it in the real world and 128 levels doesn't even begin to approach the required resolution, even with smoothing algorithms and velocity cross fading - those are not sufficient.
Kevin.
That's just... yeah, you got me alright. 
