Conference napalm::commusic_v1

    Todd's right, the problem with cymbal triggers is not memory
    but hardware - you have to do a whole bunch of adds from multiple
    places within the sample in one output conversion time.  Nowadays
    fast adders are no big deal, but they do cost a little more than
    circuits that have no adders at all.  You'd need a little hairier
    control algorithm as well, capable of keeping (OK, let's see - 16th
    notes at 160 bpm = 640 attacks/minute = about 11 attacks/second,
    so a 4 second decay means keeping about 44 pointers around and doing
    44 adds per output, and a 30 KHz conversion rate to get 15 KHz
    bandwidth necessary for good cymbals means 33 usec per conversion,
    or an add (at least 8 bits, and probably 16) in .75 usec - certainly
    possible!) well it's all back there.
    
    My ultimate cymbal box would be like a DDR-30 (let's call it a DCY-30);
    it would have the following basic cymbal sounds:
    
    	pedal closed hihat
    	sticked closed hihat
    	sticked open hihat
    	tight ride cymbal
    	open ride cymbal
    	sizzle ride cymbal
    	ride cymbal bell
    	crash

    for 8 basic samples.  It should be possible to plug in additional
    roms for more samples.  Ideally, the machine would have enough memory
    so you could take samples off 3.5" disk to configure the machine
    for a given song (so you wouldn't have to resort to program changes
    to get multiple crashes or rides within a song).
        
    Each of these would have saveable parameters that allowed you to adjust:
    
    	EQ ("brightness" or "sheen")
    	pitch
    	decay rate
    	ping/attack
    
    All samples would be velocity sensitive.  A NOTE OFF that arrived
    before the sample had decayed naturally would result in a "choke",
    except for the open hihat, which results in a pedal closed hihat.
    
    Someday, somebody's gonna do sampled cymbals right...
    
    len.
    

    How much memory though?
    
    Well, 4 seconds at 30KHz at 12 bits is 180 Kbytes/sample; that's
    1.44 Mbytes for 8 samples, or another way to look at it is 4 (count
    'em) samples per 880 Kbyte DSDD 3.5" diskette, with some space left
    over for software and other overhead.
    
    So for a real class machine we're talkin' 2 MBytes of RAM, a 3.5"
    drive, a 750 nsec 12 bit adder, a (2?) 30Khz 12 bit D/A converter(s),
    maybe some stereo ambience circuitry, control logic and microcode.  Do
    I hear $2K in production quantities?  Admittedly, this is a brute
    force approach, no attempt to encode anything more efficiently;
    maybe the guys at Roland could get it down to $1K?  8 bits companded?
    2 second samples with enveloped looping?  That would cut down the
    memory requirements by a factor of 4.  20 KHz sampling rate?
    
    But 4 samples per disk doesn't hack it!  16 would be tolerable.

    len.
    

I think $1.5K - $2K is accurate (not including pads).  But then again, the 
quality would be such that all the major recording studios would want
one.  Especially at the cost of *real* cymbals and *real* mics.  Consider 
how many *real* cymbals it would take to get the flexibility of this unit 
with its variable pitch and EG edit capabilities.  Not to mention that it
would be MIDI...

Todd. (who_sees_the_market)

Also, an adder is not necessarily needed if multiple copies of the same
cymbal voice can be active at the same time.  But in this case, it would
sound as though you had two (or more) of the same cymbal and you would
hit one if the other hadn't died out yet.  Kind of primitive, though.

The killer:
	Multiple samples of each cymbal would really be necessary for 
	different velocity levels ala piano.  This is *impossible*, given
	memory constraints.  What would really be needed is an MKS-20 full 
	of cymbal sounds rather than piano sounds.

Someone needs to create an affordable box that can "Synthesize" cymbal
sounds (easier said than done).  That would get rid of the memory requirement
problem...

Todd.

    If you could get in on a "pad" then the electronic drummer of tomorrow
    could be totally electronic, eliminating the need for acoustic cymbols
    think of the boon to headphone type rehersals, both for bands and
    the parents of budding young drummers :-), since it would be midi
    then the pad approach should work well...and I could slave it to
    my 707.....                                          
    
    dave (anxiously awaiting the product announcement)

Of course the only way of getting a flawless MIDI cymbal would be to put
a cymbal on a small stand in a small acoustically insulated box that houses
a microphone, a mechanical striker mechanism, and a computer that acts as
a "MIDI to striker" converter that strikes the cymbal at a velocity 
corresponding to the velocity imbedded in the MIDI info...

Todd. (Now_how_do_we_get_MIDI_vocals?)

    Given the state of readiness of other products shown in public,
    maybe we should reserve some space for the Winter NAMM show :-).
    
    		Derek

>...and a computer that acts as
>a "MIDI to striker" converter that strikes the cymbal at a velocity 
>corresponding to the velocity imbedded in the MIDI info...

	You probably said this in jest, Todd, but I've 
	*actually seen* such a beast!

	There is a company (whose name escapes me now) which
	makes these elaborate arrays of solenoids, which clamp
	on to conventional instruments and are controlled by micros. 

	They have one for piano: yep, 88 little rubber-tipped
	rods, mounted on a slab of plexiglas, clamped above the
	keyboard.

	They also have a "percussion set" which has a set of cymbals,
	snare drums, and other percussive things, all mounted on
	stands, and with these solenoid-gizmos attached.

	The solenoids are velocity-sensitive, of course (that is,
	they can be programmed to strike at different forces).

	They had a demo booth at last year's International Computer
	Music Conference (ICMC) in Vancouver. Big surprise: the
	Yamaha, E-mu, and Kurzweil booths (the latter complete with 
	Bob Moog) drew *much* more interest!

							Scot

	p.s. I might still have their literature, if anyone is
	     interested. I seem to recall they are out here in
	     Massachusetts.

    Hi tech! Seem to remember some mutant beast of a
    player piano that had all sorts of percussion that
    could be activated from the roll. Sort of a Juke box.
    Why not just dig one of these up, add midi, and mike it?
    A small micro, Uart, and B-I-G power supply should do it.
    
    Firmware anyone?
    
    Should we leak it now , or wait for a formal anouncement?
    
    Ron (who-gave-up-long-ago-on-a-good-analog-cymbal-patch)

> Hi tech! Seem to remember some mutant beast of a
> player piano that had all sorts of percussion that
> could be activated from the roll. Sort of a Juke box.

	Yes, "beast" is a polite description of these machines.
	They were the music box makers' last-gasp attempt to
	compete with the phonograph, so long ago.

	These multi-instrument machines made more noise than
	music, but are still fun to watch. There was one in 
	the Press Club restaraunt in Lowell a few years ago,
	wonder if it's still there?

						Scot

    The piano version of such a thing is called a "vorsetzer".  They've
    been around for a while.  Older ones were driven off piano rolls,
    and the newer versions off cassette tapes.  I think Sony and Marantz
    collaborated on one a few years back (or maybe they each had their
    own version).  I have also seen photographs of a similar contraption
    that "played" the violin, and have often considered designing one
    for a guitar.
    
    re Tom's estimate of costs - parts cost is far and away the smallest
    component of a modern electronic instrument.  Have you considered
    packaging, power supplies, assembly, sales and field support costs,
    documentation, inventory, etc.?  My price estimate was based on
    going rates for currently available instruments of similar complexity.
    And in any case, the circuitry to integrate the basic components
    (RAM, convertors, disk drive, etc.) will at least triple the component
    count.
    
    What surprises me most is I know of *nobody* working on such a box.
    I would buy one almost instantly.  Maybe I should write a product
    proposal to Roland...
    
    len.
    

    If (well, that's a foregone conclusion) and when I acquire a decent
    sampler, I plan to try essentially just what Edd suggested back
    a ways.  A sampler that would hold 16 2-second samples (e.g., a
    Roland S50 or a Sequential 2002) could be set up so 16 identical ride
    cymbal samples were each assigned its own note number.  Then by
    appropriately programming the ride beat (it would look like a repeated
    scale) you could invoke successive samples so each could play out
    almost its whole length before being retriggered. A 120 bpm 8th
    note ride entails 4 events per second, so 8 2-second samples would
    suffice, or 16 4-second samples, without truncating any decays.
    16 4-second samples is beyond the means of most samplers today.
    
    len.
    

    What if.....
    
    2 samples were loaded into memory. Sample one would be a simple
    hit on the ride. Sample 2 would also be a hit on the ride, differing
    from the first in that it would be a sample of any other hit on
    the ride other then the first, hence it would contain decaying
    previous hit envelopes. A software pointer positions it self on
    the 1st sample if no other ride sample has been played within the
    decay time, or sample 2 if sample 1 is still decaying. Sample 2
    could be re-triggered over and over.... 
    
    Yes? or "go back to sleep, Edd"?
    
    Edd

    I suppose you could do this, but I think all you're doing is slightly
    reducing the amount of truncation.  Consider the 3rd hit - it truncates
    the second one, but since the second one starts later, it just
    truncates it later in the decay.  Even if the second sample had
    "ten hits worth of decay" (great name for a band, what?) built into
    it, it would still get prematurely truncated (another great name?)
    by the next hit.  So you're just changing the sound at the truncation
    boundary to a differnt sound.  It would probably work better than
    what's done today, but it also might sound weird with the second hit
    suddenly introducing a couple of beats' worth of cymbal decay!
    
    Nice try, but I don't think it cuts it.  Maybe a more elaborate
    version, but at some point you wonder if the extra mechanism to
    simplify things is really a simplification...  It's like a short
    cut that takes more time to decide to use than the shortcut saves.
    I've seen a lot of "clever" programming tricks that turn out to
    be this sort of "improvement".

    len (who reminds you to beware of "experts" who say "it won't work")

    re .10, .11
    There were things called "Nickleodeons" or "Band organs" that carried
    this to an extreme.  Some of them had whole brass & reed sections
    as well as pianos & drums in them, all driven from a paper roll.
    
    Where to see one ?  The last two I remember seeing are:  In the
    depot at the Edaville Railroad excursion line, and in the Circus
    Museum in Baraboo, Wisconsin.   Some of these even had Steam Calliopes
    in them !  Now that's an interesting timbre to try to synthesize !
    You really have to hear one in person to appreciate what they sound
    like.  I guess not many people have these days.
    
    In general: The addition of a second Ride & Crash cymbal would make
    a great addition to my TR505.  Everyone says the ride sound is good
    for a cheap drum machine, but there's no way that it can sound like
    a real ride cymbal because it always decays completely before the
    next attack. 

    
    Ok, so I've had a glass of wine, but consider:
    
    1 sample of cymbal ride strike....and  multiple pointers.
    For simplicity, say pointers = 2.
    
    First strike is Simple Sample (ah, another band name anyone?).
    Another strike starts another pointer from the sample top.
    It's sample value is added (maybe scaled too) to the value 
    of wherever the first is pointing to as it continues cycling
    through the waveform file. They go on to completion.
    
   There. You have a complex wave that represents 2 strikes.
   Isnt that what you are "really" (read: sort of) doing when you
    add a strike to a decaying strike?
    
    The overhead is the addition of N pointer samples. If a sample
    pointer reaches the "waveform file end". It can be used for the
    next strike, so N can be small. (The analogy is a round-robin
    voice assignment algorithm). Let's try it Len!
    
    Oh, when I worked at ARP (ah,RIP), we used (get this!) a SWAG of
     10*parts_cost = average_user_discounted_price

    That seemed to work fairly well for estimating new designs that
    you were trying to get out the door. 
    
	Ron (reminising-after-only-one-glass-of-wine)
    

    Yeah, what you're proposing is basically the same thing I did, but
    with fewer pointers.  The maximum number of pointers you need is
    the sample length divided by the time between events; any more pointers
    won't buy you anything.  The minimum number of pointers is 1, which
    is the way most things are built today.  Pointers are cheap; if
    you have the time, you can do the adds serially, and once you have
    two pointers you need the adder; unless you don't have enough time,
    you don't need an additional adder per pointer, so you might as
    well use as many pointers as you have time to add in the interval
    between output conversions (at 30KHz, that's 33.3 usec.).  A 12-bit
    parallel adder is cheap, and can probably do the add in around .5
    usec. max, so it's reasonable to consider having as many as 60 or
    so pointers!  The additional logic to support a lot of pointers
    is relatively trivial, so it's probably worth going all the way.
    
    len.
    

    I was just guessing, and I said "max".  We clearly don't need ECL
    for this.  Something like low power Schottky should suffice.  I've
    fogotten what gate delays are like these days so just guesstimated
    50 nsec and 10 gates as worst case.  I'm a software guy, gimme a
    break.
    
    len.
    

    (My apologies to those of you who may have received this by mail.
    I couldn'y reach COMMUSIC:: earlier.)
                                 
                               *    *    *
    
    What do you think of this idea as a possible solution to the age-old
    ride cymbal sample problem...
    
    (Assume Mirage. It's the only sampler I own.)

    Let's say we want to play an 8th note ride pattern...
    Split the keyboard into 3 zones...

                 Zone 1.  A single hit that will last at least as
                          long as a 1/4 note. (I think looping would
                          be all but impossible, but at 120 BPM a 1/4
                          note would only require .5 seconds, well within
                          the capabilities of the Mirage.) Assign this
                          to be note number n1.

                 Zone 2.  Also a single 1/4 note hit, but this one is taken
                          from the middle of a series. Assign this to be note
                          number n2.

                 Zone 3.  Identical to Zone 2. Assign to note n3.



    We would play one note from zone 1 for a quarter note. Overdubbed (merged)
    onto the same measure would be an 1/8 rest followed by 4 1/4 notes from zone
    2.

    A second merge would consist of a 1/4 rest and 3 1/4 notes from zone 3.
	
    *:== Note on
    ...:== release (which is what cymbals do the most of)


    Beat              1   &   2   &   3   &   4   & . . . . 
    Zone 1 (n1)       *.......
    Zone 2 (n2)       R   *.......*.......*.......*
    Zone 3 (n3)       R   R   *.......*.......*......
                      _______________________________
    8th notes!!       *   *   *   *   *   *   *   * .....

                                 (OK, the end is a bit ragged....)

	But, what we do get is attack portion of one sample being played at
        the same time as the release portion of the previous one.
        More like real cymbals, yes?

        Perfect? No. Better? Well, I'm seeking opinions....
         
    	Does anyone have a very expensive ride cymbal I can borrow? (Thanks
        Todd, but no thanks, yours are all cracked.  :^))

    Edd        

    Well, sounds good. I think I see the need for a 'starter' cymbal hit, 
    as well as a middle hit, and you just go 12323232 12323232 etc.
    
    However, on my Emax, using 'nontranspose' mode (1 sample triggered
    by multiple keys) I get a very realistic ride cymbal without setting
    up a 'starter' cymbal hit.. As the sample lasts ~1.5 secs before
    dying away, I've got 8 keys assigned to the same sample.. remember
    that the unit has 8-note polyphony.. So I just play 12345678 12345678 in
    meter.. allowing all hits to die away naturally before being
    retriggered. I've got velocity assigned to volume and filter, the 
    harder the hit the louder and brighter the result.. just like real 
    drums, right ?
    
    kmtwo
    
    

    > I think I see the need for a 'starter' cymbal hit...
    
    Right! Me too!! That's what 'Zone 1' was for.
    
    Edd

    Does this really work, or can one hear a little 'popping' as each
    sample kicks in?

    I've got better results by actually triggering the same sample
    again each time and hence NOT have the decay of previous
    samples sound. Since it sounds clearer this way I find it
    more usable in a musical context.
    
    Actually, I think it's just that what happens when you hit
    a real ride symbal many times in sequence. ==> the earlier decay
    get's cut by the next hit. 
    
    Timo
    

    re: .24   I dunno, I haven't tried it yet, but why to you think
              there would be a 'pop'?
    
    re: .25   I think our resident rhythmically arrested boom-slam
              expert would disagree.
    
    Edd

    re -.1: 'pop' suspected because of a transient that occurs as a
    result of the mismatch of samples, i.e.; the signal at the end of the 
    previous sample may not exactly match the signal at the beginning 
    of the next sample.  Of course, I'd expect a low-pass filter to smooth
    such a transient out.  Might also be a mismatch in amplitude that
    would be noticeable.

    I suspect this would not be a problem as the attack of the hit being
    played would effectively mask any level mismatch between samples.
    
    I imagine at one tempo, the level would match perfectly....
    
    ...seems the only thing to do is actually try it! (What? you mean
       people can actually *play* these things? Not just collect 'em?)
    
    Edd

    
    sounds good to me.
    
    (was it) Timo (?), point was that 'real' repeated cymbal
    hits DONT cut off and retrigger the sound again,eh?

    I want a midi-to-robot-arm-to-Zyldzian converter.

    yet another project.
    

    I understood Timo to be saying that on a real cymbal, subsequent
    attacks suppress the decays in progress.  I think that's not true,
    and can imagine no physical mechanism whereby this would happen.
    Also, ride cymbals are the least convincing of any synthetic drum 
    sounds I have access to, and I believe this is reason.  Having three
    ride cymbals whose sounds I treasure, I speak from a lot of experience
    comparing real cymbals to synthetic ones.
    
    Ah, for some real data...
    
    len.
    

    I hate to dissagree with a real Boom-Slam expert, but this
    keeps bothering me...
    
    If you have a cymbal vibrating and you touch it with your
    finger, it will affect the vibration slightly and the sound
    will perhaps decay a little faster. Well, how about hitting it 
    with a stick. I think it will have an even greater effect on the 
    ongoing decay - and even without taking into account the fact that 
    some of the newly generated vibrations are bound to be out of phase 
    with the earlier ones and will therefore have a suppressing effect.
    
    I'm not saying (at least didn't mean) that the earlier decay
    will be cut altogether. Only that it's much more complex and 
    that triggering the same sample again sounds closer to the 
    real situation.
    
    Think of a real player (rhythmically arrested one) hitting
    8 separate real ride cymbals in sequence. I'm sure it will
    sound very different from what our ears are used to hearing 
    when using only one cymbal for the series of hits.
    
    Timo                               

    BUT!!!!
    
    By sampling a subsequent 'hit' we should be able to capture any
    nuances and weirdisms *whatever* they are.
    
    I'm gonna do it.
    
    Edd

    Touching a cymbal with your finger is very different from striking
    it with a stick - the former damps the vibration (takes energy out
    of the cymbal), the latter further excites it (adds energy to the
    cymbal).  While it is possible that the subsequent excitation might
    be "out of phase" with the the ongoing vibration, the waveform of
    a cymbal is so complex and inharmonmic that it seems rather unlikely
    that this phase cancellation could amount to anything.
    
    The basic question is whether or not a cymbal behaves linearly with
    respect to excitation (i.e., whether the response to a sum of inputs
    is the sum of the responses to the individual inputs).  I would
    not expect a cymbal to be perfectly linear (very few real physical
    systems are), but I'd expect it to be more linear than not, at least
    up to some amplitude limit (where the cymbal "saturates" or worse,
    breaks).                  
    
    The "multiple cymbals" thought experiment is interesting but can't
    be performed, because no two cymbals sound alike.  Other spatial
    effects would also intrude, as well as the problem of getting
    consistent strikes across multiple cymbals.
    
    Be that as it may, it seems clear that subsequent strikes do not
    utterly truncate the ongoing vibration, as happens on synthetic
    cymbals.  Something substantial carries through, you can hear it
    build up very clearly on a real cymbal.  When carried to extremes
    (e.g., a 16th note roll on a cymbal) this "buildup" comes to dominate
    the sound.
    
    len.
    

    RE: -1
    
    > Touching a cymbal with your finger is very different from striking
    > it with a stick - the former damps the vibration (takes energy out
    > of the cymbal), the latter further excites it (adds energy to the
    > cymbal).
    
    I was thinking that the latter (striking it with a stick) would
    BOTH damp and add new energy to the cymbal.
    
    > The "multiple cymbals" thought experiment is interesting but can't
    > be performed, because no two cymbals sound alike.  Other spatial
    > effects would also intrude, as well as the problem of getting
    > consistent strikes across multiple cymbals.
    
    I was thinking of it as a hypothetical experiment, or more as
    a way of thinking or picturing what the "buildup" might sound
    on a linear system (to put it on your terms).  
     
    > Be that as it may, it seems clear that subsequent strikes do not
    > utterly truncate the ongoing vibration, as happens on synthetic
    > cymbals.  Something substantial carries through, you can hear it
    > build up very clearly on a real cymbal.  When carried to extremes
    > (e.g., a 16th note roll on a cymbal) this "buildup" comes to dominate
    > the sound.

    I agree.
    
    
    RE: -2
    
    Sampling a subsequent hit sounds like the right way of doing it.
    However, if you have the decay of the previous trigger still going
    when "a subsequent sample" is triggered, doesn't this mean that
    you get a double decay for the previous hit since you have it 
    in your sample as well. 
    
    Either way, keep us posted how it works out (sounds).
    
    Timo
    

    I did think of a big bugg...
    
    What we need is clock-synced envelope. My idea is only gonna be
    'accurate' at *one* tempo. Any other tempo will cause the levels
    between the 'decay' at the end of one sample and the 'decay' at
    the beginning of the subsequent sample (even if they are the same)
    to be off. I'm not performing anywhere. This may not be a serious
    problem given the probable masking caused by the hit at the beggining
    of the sample.
    
    Edd   
    
    

    Not so serious, Edd.
    
    Your idea works FINE as long as a key triggering a given sample
    is not pressed again until its sample has run its course (timewise).
    
    karl

    Last night I attempted to take the first steps in seeing if this
    will work. It is not gonna be as easy as it initially seemed. (But
    what ever is?)
    
    Len Fehskens provided me with a very clean sounding tape of his
    Avedis Zildjians. We warned me that despite the *apparent* levels
    of the cymbals, the VU meters indicated that the signals was down
    under -20 db. Sure enough, when I got how the tape was almost
    gratingly loud, yet the meters nary budged. (Let me emphatically
    state that this is NOT a reflection of Len's recording technique.)
    
    Anywho, armed with this tape, I attempted to load a sample into
    the Mirage. My method of doing this consists of running the stereo
    deck into 2 channels on the board and then taking a mono signal
    from the headphone PFL jack. Always works fine....
    
    ...except this time. Due to the low levels, it was impossible to
    find a reasonable compromise between output level and sampling
    threshold. Any threshold setting low enough to capture the cymbal
    was also low enough to be activated by tape/board noise. Any attempts
    at raising the threshold would result in "No signal".
    
    What next? We try loading a real live cymbal in with a mike and
    a compressor. Stay tuned to this station for all the exciting details.
    
    Edd