[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference napalm::commusic_v1

Title:* * Computer Music, MIDI, and Related Topics * *
Notice:Conference has been write-locked. Use new version.
Moderator:DYPSS1::SCHAFER
Created:Thu Feb 20 1986
Last Modified:Mon Aug 29 1994
Last Successful Update:Fri Jun 06 1997
Number of topics:2852
Total number of notes:33157

440.0. "Roland MKS20 Digital Piano - Not Realistic" by APOLLO::RANTALA () Mon Jul 21 1986 14:33

I listened to the Roland MKS-20 over the weekend at the Wurlitzer
in Worcester. The most impressive feature was the velocity encoding.
The harmonic content of a note that was played fortissimo was very
different than when played softly. The result was quite breathtaking.
So it plays a mean one-note piano (I hear that stuff all the time in
the dentist's office). 

The bubble burst when I played a chord. It didn't "sing". It faithfully
reproduced all of the notes, but the notes were isolated from each
other. There was no interaction. The difference between a major and
a minor chord just begins with the flatted third. An acoustic instrument
forces interaction between notes because they are all attached to the same
bridge and soundboard. The waveform of a string that is played separately
is different than the waveform it produces when it is vibrating along
with another string.

Each note of the MKS-20 has 128 samples of the corresponding Steinway note,
with the variable being the velocity at which the hammer stikes. I imagine
that an astronomical amount of information would have to be stored in order
to include any kind of attempt at the "interaction" variable in every 
note's makeup. 

I think that Roland has done a great job for the money, but piano digitization
still has a way to go. Has any manufacturer made an attempt to make the notes
talk to each other?
 

				Joe

T.RTitleUserPersonal
Name
DateLines
440.1CANYON::MOELLERPins in my Software dollsMon Jul 21 1986 14:407
    There is already a note on the Roland MKS-20. I run mine thru a
    Alesis MIDIverb, which gives it the appropriate ringing overtone
    exchange. It isn't perfect, but it's close enough to fool anyone
    once it's recorded. Check out my 'Is it Real or Is It Roland Challenge'
    in the MKS- note.
    
    karl moeller
440.2RE: 128 samples per noteBARNUM::RHODESMon Jul 21 1986 15:563
    I thought it used "Structured Adaptive Systhesis" rather than sampling.
    Or is SAS an algorithm that selects and conditions samples?
    
440.3dunnoAPOLLO::RANTALAMon Jul 21 1986 18:351
    I was only paraphrasing the salesguy (very dangerous, I know).
440.4No Buddy NoseERLANG::FEHSKENSTue Jul 22 1986 11:169
    I have yet to see a clear description of just what SAS really is.
    
    128 samples per note (baldly) seems to me impractical, even with
    256K ROMs the cost would be unmanageable (think about the amount
    of storage required).  SAS has to be some kind of algorithmic
    trick.
    
    len the guesser.
    
440.5SAS is additive synthesis + ???PIXEL::COHENRichard CohenTue Jul 22 1986 11:208
    According to Bob Moog in Byte magazine, SAS is a resynthesis technique
    mostly involving additive sine wave synthesis, ala the Crumar GDS
    (which I did some development work on at Bell Labs), Digital Keyboards
    (also Crumar) Synergy, etc. Wendy Carlos is a big proponent of additive
    synthesis. The new Kurzweil 150 is using a similar technology.

    	- Rick
    
440.6Wave Bye Bye, Now, Nice EnvelopeERLANG::FEHSKENSWed Jul 23 1986 11:0156
    I read the Keyboard review of the MKS-20 last night, and it confirms
    the notion that there's a sample for each note for each velocity
    level.  This is a staggering amount of information, even if the
    samples are only a single cycle long.  I put on my thinking cap
    and figured out just how much storage was required for one cycle's
    worth of 128 (MIDI notes) * 128 (velocity values) samples.
    
    First some assumptions - assume the sampling rate is 40 KHz, and
    the samples are 12 bits.  Assume each sample uses only as much memory
    as it must.  128 notes is 7 octaves, and each successive half step
    up the scale requires 1/(12th root of 2) as much storage as the
    previous.  We could do this as an enormous power series, i.e.,
    
    	1 + 1/(2^(1/12)) + 1/((2^(1/12))^2) + 1/((2^(1/12))^3) + ...
    		+ 1/((2^(1/12))^127)
    
    As this was a "back of the envelope" calculation, my Amiga was busy
    plotting yet another detail of the Mandelbrot set, and I couldn't
    find a closed form for such a series, I resorted to some simplifying
    approximations.
    
    First, each octave takes up half as much space as the next lower
    octave.  This reduces it to a problem of evaluating that series
    only to the  1/((2^(1/12))^11) term.  The whole set then takes
    
    	1 + 1/2 + 1/4 + 1/8 + 1/16 + 1/32 + 1/64
    
    times the storage for the first octave.  Well, this is easy - that's
    just a factor of 2 (actually, 2 - 1/64, but between friends, what's
    1/64?).  Now how do we get the storage for one octave?  Now we get
    sweeping - a reasonable first approximation is that the *average*
    sample length in the first octave is somewhere between 1 and 1/2
    times the lowest note's sample length.  Pick 3/4 as a conservative
    average value. 12 * 3/4 = 9, so say the first octave takes (less
    than) 9 times as much storage as the first note.  All 7 octaves take
    (less than) twice that, so the whole sample set FOR ONE VELOCITY
    VALUE takes 18 times as much storage as the lowest note does.
    
    So how much storage does the lowest note take?  MIDI note 0 is the
    16 cycle C (it's not exactly 16 Hz, but that's close enough).  At
    16 Hz one cycle sampled at 40KHz requires 40K/16 samples, or 2500
    samples of 12 bits each.  That's 3750 bytes.
    
    So, that comes to 18 * 3750 or 67500 bytes per velocity value. 
    If there really is a sample set per velocity value, we need 128
    * 67500, or 8640000 bytes.  Now, 8 megabytes is an awful lot
    of storage, even if it is ROM, to cram into a $1500 2 high rack
    mount box.  And even if there were that much ROM, you still need
    to do a lot of processing to make it sound good - a piano most
    assuredly does not owe its magnificent sound to a single waveform
    repeated throughout a note's life.
    
    What this all says to me is unless/until Roland spells out exactly what
    SAS is, we (and the magazine editors) are just waving our hands.

    len (who isn't that good at arithmetic, so check my work)
440.7my god, what have I done?APOLLO::DEHAHNWed Jul 23 1986 12:329
    
    
    8 megabytes of ROM???
    
    It would take forever to read that out even at 200ns/word.
    
    CdH
    
    
440.8...and the mystery continues...BAILEY::RHODESWed Jul 23 1986 14:3317
    Nice work Len.
    
    No way there are 128 samples per note.  No way.  The Kurzweil can't
    even do that.
    
    I don't think SAS has anything to do with "sampling".  My reason
    for saying this is their line of Juno synthesizers sport this
    same technology.  As far as I know, the Junos allow the user
    to create custom sounds (like the DX series) which couldn't occur
    if the technology were based on samples (unless it were like the
    DW series from Korg, which I doubt).
    
    What we need is to sneak a peak at a Juno manual.  Are you gonna
    be going down to EU's any time soon Len?
    
    Todd.
    
440.9FM pianoTROLL::DIETRICHFri Aug 01 1986 19:202
    Has anyone heard of frequency modulation?
    
440.10???DECWET::MITCHELLFri Aug 01 1986 20:214
    Yes, but why do you ask?
    
    
    John M.