What is the Organic Drum Machine? The first metronome may have been a pendulum like Cuthbert Calculus’ (above); gravity driven pendulums were occasionally used for musical timing by choirmasters and others (including Bartok) starting about 1700. People have used geared machines as musical timekeepers for 200 years. This has become so common that we sometimes confuse clock time and musical time. In the past few decades drum machines have become useful, aesthetically appealing, and pervasive in many styles of music. How can technology, science, and music work together to build a drum machine that uses human time rather than clock time? The Organic Drum Machine, or ODM, is intended to be a new kind of rhythm machine. Like most drum machines the ODM’s primary function will be executing rhythmic sequences, especially the kind of layered and syncopated patterns found in groove-based music. The main difference between the ODM and currently available rhythm machines will be in how timing data is compiled and parsed. Most drum machines use clock time to regulate their sequences. In these machines 60 beats per minute means exactly that. The ODM will use algorithms and statistical models based on hard data and data-driven generalities about human periodic behavior, from musical timekeeping to respiration and beyond, to determine timing. In the ODM units like the minute will have little or no direct bearing on a sounded result and 60 beats per minute will be a starting point rather than a strict measure. This latest “Funky on the One” post is an attempt to more clearly define what the ODM is and relate its proposed functionality to my earlier posts on the topic.
My first bass teacher was Darren Solomon, and his first substantive assignment involved the metronome. “Put the metronome on very slow,” said Darren, “and play right with the click. Work on this a lot: it’s the key to being a good bass player.” And I did. First at 60 BPM, then 50, then 40, then 36, and slower, I would sit and play low Cs as right on the click as I could. It’s pretty hard to do! And it’s great for development of accurate timekeeping. Thus I and thousands of others have been introduced to mechanical time, and learned to equate it with accurate musical time.
Like many changes in music over the past two hundred years or so (the familiar mechanical metronome in common use dates to around 1816) it’s easy to forget what things must have been like before the metronome existed. Just as recording and amplification created immense changes in the portability, amplitude, and timbre of music, metronomic time created tempo external to human musical timekeepers. Along with this paradigmatic shift comes a transfer of agency–tempo in the mechanized age ceases to be an internalized feel related to people and bodies and begins to be indexed relative to beats per minute, MIDI ticks, or samples per second. Thinking about musical time is transformed completely by this shift from relative, subjective, and embodied to absolute, objective, and mechanical. While few musicians would say that perfect mechanical time is a musical ideal, and some great players eschew the metronome completely, there is no doubt that, especially since drum machines have become common use items, the ability to keep time with clocklike accuracy is a desirable skill among trained musicians, and music with a mechanical beat has become standard in many styles.
Imagine if Darren had given me the same assignment, but the goal was to play with someone else at these slow periodicities. The skills involved would be very different, and the results different too. Instead of an infallible metronome click accompanied by my variously inaccurate low Cs, both players would be playing low Cs at slightly varying tempi, each working to match the other. Issues alien to clock time would come into play-who is “right?” What is the desired tempo anyway? Why bother with this? Indeed Darren’s assignment, and much of my and many others’ training in musical time, depends completely on availability of and faith in a mechanical timekeeper.
While clock-based timing can be effective for practice, analysis, and in performance or recording contexts, this clock time is but a recently developed approximation of human musical timekeeping processes that are related to the wealth of time sensitive biological, social, and environmental activities humans and all living things undertake constantly and at varying periodicities (respiration, walking, digestion, waking/sleeping, agricultural cycles, etc). It’s easy to confuse “solid” steady timekeeping with mechanical time, but I propose that human enactment of rhythm is fundamentally different, unrelated to clock time, and at deeper levels of analysis human musical rhythm will reveal as yet unquantified types of accuracy and patterns at micro and macro timing hierarchies. New information about this aspect of human rhythm will lead to a better understanding of rhythmic expressivity.
Effective and unique rhythmic expression, even when it is “just” a well-played pulse, brings a great deal of nuance and meaning to music, yet the exact quality of such nuances and meanings are largely unavailable to the composer, music analyst, and programmer. How can we tap into this deep well of important musical information? The Organic Drum Machine (ODM) project is my proposed path towards digital rhythm derived from human rather than mechanical data sets. My main goal is to create a rhythm machine using Max/MSP/Jitter that can execute onsets using data and formulas derived from human timekeeping models. Some of these as-yet-to-be-built models will include concepts discussed in earlier posts such as metronomization, preparatory microtiming, hypermetronomic information, centeredness, and wet vs dry musical processes. I have yet to research other human timekeeping paradigms, especially long-memory statistical models, and further research is required on all fronts.
At root this project is about expanding and improving the computer rhythm machine. The ODM is intended to be used in compositional and performative contexts; it’s a creative tool. As a musician, composer, and music lover the rhythmic patterns found in African diasporic music have been a source of great inspiration and joy for me, and many musicians in these traditions from the past 30 years or so have used computerized rhythm in their compositional, performative, or recording practice. Indeed certain trends and periods in music production have been defined by the sound of specific rhythm machines (Roland’s TR 808 being a primary example), and it is hard to overestimate the impact computer rhythm machines have had on the timbre, structure, and production of many musical styles. The time has come for rhythm machines to move beyond clock time and into human time. Such a shift in focus at the programming level will invite new thinking about how, when, and why we use computerized rhythmic sequences. We are long past the time when computers had the capacity to emulate rhythm using human models of timekeeping, and are approaching a moment where computer music and human performance can be fully integrated. Clock time is no longer a sufficient tool for measurement and expression of musical rhythm.
So what is the ODM and how does it fit in? At this point I am not entirely sure. While I envision a future program that uses many algorithms to generate novel, interactive, highly syncopated, and timbrally heterogeneous rhythmic textures related to the Brazilian, West African, and Cuban percussive traditions I enjoy, my initial plans are not nearly this ambitious. I hope to expand the functionality, nuance, and temporal palette of rhythm machines starting with the “simplest” of concepts: a steady pulse. Computer generation of a steady beat that is not clock based is not to be trifled with. How do we do it? How and why does a person clapping a pulse deviate from clock time, and, if they are clapping along with someone else, how will their patterns differ from and influence each other? What happens when two percussionists play even a basic interlocking groove? These questions are remarkably deep, and while I do not plan to solve them in a definite way, the ODM project addresses such issues.
As I have worked on these posts I have made some decisions about what the early generations of the ODM machine should accomplish:
1) Maintain a “steady pulse” without adhering to clock time.
Two analyses of improvisational microtiming (Benadon 2009, 2007) and a computer emulation of grooving rhythms (Wright and Berdahl, 2006) tended to focus on sub-tactus onset events. In these studies the tactus was either assumed to be clocklike (Benadon 2009, 2007) or a mechanical click was utilized (Wright and Berdahl 2007). These studies sought to parse, contextualize, or emulate expressive microtiming at the sub-tactus level by necessarily ignoring the fact that any performance not utilizing mechanical time will include temporal changes (speeding up, slowing down, changes in emphasis, etc) at the tactus, metric, phrase, and section levels.
The ODM, in contrast, is meant to generate onset events using human-like timing data at every level of a rhythmic hierarchy, and as such a first step is to develop algorithms that execute a periodicity without needing to return to unison with some clock-based timing unit.
Most rhythm machines have various “swing” and randomness settings that can be used to vary the onset events of a particular programmed sequence. These settings function by departing from then ultimately returning to a fixed mechanical time, for instance quarter note = 60 BPM. The ODM will be fundamentally different from other rhythm machines. For example, it is possible in many rhythm machines to set up an eight measure rhythmic pattern in 4/4 time with a metronome marking of quarter note = 60 BPM that uses swing and randomness settings to generate a sequence of unevenly (“expressively”) executed interlocking or overlapping patterns. However, this sequence will last exactly 32 seconds no matter how extreme the swing and randomness settings may be. With the ODM such patterns will last about 32 seconds, but the timing information will, like human performance, “naturally” speed up, slow down, and never have to snap back to a clock based temporal grid. At first I plan to err on the side of inaccuracy such that onset events can be made to sound more ragged than a human timekeeper, and work towards varying levels and kinds of human accuracy. This first step of maintaining a steady pulse will be followed by efforts to emulate some kind of basic two part rhythmic pattern such as clave and tumbao, walking bass and high-hat on 2&4, etc (see #3 below).
2) Execute rhythms for extended periods that vary in ways similar to human rhythmic variance (for example, as found in musical timekeeping, stride, and pulse rate cycles).
Studies of human gait dynamics reveal that “long-term fractal dynamics of the stride interval are normally quite robust, they are apparently intrinsic to the locomotor system, and they exist at a wide range of gait speeds” (Hausdorff 2007, 561). Self-similar gait patterns repeat over very long periodicities into thousands of strides (Hausdorff 2007, 560). Similar fractal patterns and scaling can be found in human respiration (Peng et al. 2002). In the fields of biomechanics, medical dynamics, and music, variation from a mean rate (be it a rate of pace, pulse, or musical rhythm) can easily be misidentified as statistical noise. In many fields this noise is being reevaluated and considered a key aspect of biomechanical function. How can this new data be incorporated into materials for creative musical work? In scientific terms maintaining a musical pulse is “isochronous serial interval production” (ISIP), and recent studies suggest, as I do, that “ISIP is a more complex process than is assumed by influential timing models and theories, and that realistic modeling of human timing must account for nonlinear variability patterns” (Madison 2004, 105).
While best practices and data are still to be determined, the ODM should generate note onsets using various human timekeeping algorithms, emulating the long term patterns found in phenomenon such as walking, breathing, and musical timekeeping. It is common for rhythm section players in groove based music to play thousands of notes in a single piece, and the data that dictates the ODM’s note onset events will reflect the ways in which we maintain rhythm over such large numbers of events. Will such a shift yield a perceivable effect for music using machine rhythm and drum machines? How can such a change influence composition and rhythm programming?
3) Execute predetermined two or three part syncopated patterns, each part of which is buffered from the other parts, and maintain these rhythms in a human way.
This task requires further development of algorithms that emulate human timekeeping tropes. In a 2001 experiment Yanqing Chen et al. examined ISIP intervals for participants tapping a computer key in phase with and syncopated against a sequence of mechanical beeps. As with Guy Madison’s 2004 study, the results showed a clear pattern of long-memory influence on ISIP intervals and a successful fractal modeling of the phenomenon (Chen et. al. 2001, 4). In addition, Chen’s experiment revealed that the individuals studied showed a longer statistical memory in their syncopated tapping patterns. In other words, tapping along with a pulse and tapping syncopated against a pulse involve different methods of timekeeping. Theoretically, statistical models could be built that reflect how people maintain various kinds of interlocking patterns to form grooves, and these models could be used to supply timing information for a rhythm machine such that patterns that use triplets involve triplet syncopation statistics, patterns based on 2:3 clave will use 2:3 clave based statistics, etc.
Entrainment is both a biological and creative phenomenon; we are entrained by circadian rhythms and become entrained by effective musical rhythms. What is less clear is how the diverse multilayered hierarchical rhythm styles found in music affect our brains differently, and how musicians in various traditions use their brains differently to keep time (become actively entrained) in interlocking rhythmic patterns. One thing is sure: musicians have to listen to each other to make their rhythms fit together properly. This involves coordination of internal processes with ongoing auditory input. Various studies (for an overview see Janata and Grafton, 2003) have shown that metrical grouping and familiarity with the rhythmic traditions being referenced are key to a musician’s ability to process multipart rhythmic structures. While it’s probable that musicians in an ensemble share a common sense of metrical grouping and style, each musician playing a groove must also, while keeping their pattern, adjust to their collaborators’ patterns. The ODM’s architecture should incorporate aspects of this process, and its timing adjustments will necessarily be buffered at various intervals so that these adjustments are not too quick. As a bassist I sometimes feel that I focus primarily on my part, assuming that my internal clock and those of my fellow musicians are well adjusted enough to be in sync. It’s only when something notable happens (someone is drifting temporally, or inviting interaction) that I adjust my own playing. I imagine a similar programming attitude; when executing a multipart rhythm, the ODM should selectively “listen” and only adjust when a groove’s onset intervals have crossed some threshold of accuracy. Ideally this will lead to mechanized rhythms that reflect our own processes of entrainment as performers, and yield grooving rhythms with a fresh sound.
4) Accent beats both with intention (in the “right” place) and human-like inaccuracy (a little louder here, a little softer there).
Accents can be created by patterns of emphasis and de-emphasis within a musical texture or supplied by a listener based on their own expectations (Grahn and Brett, 2007). They perform a key function in hierarchical grooving rhythms; various on- and off-beat accents allow listeners and performers to find the unique “pocket” central to a rhythmic pattern. While initial models of the ODM will not be programmed to algorithmically generate new rhythmic structures and accent patterns, this is a future goal, and towards that end some gestalt for organizing accents should be developed. The ODM’s initial sequences will be based on familiar patterns such as the Brazilian maracatu or Cuban guaguanco, and beats will be accented using the tropes standard to these traditions. I wonder if a systematic study of diverse hierarchical rhythms might reveal some common accent locations and ways of emphasizing syncopations. Such data could be used to build algorithms that allow future iterations of the ODM to accent certain beats appropriately (i.e. in a culturally informed human way) as it generates novel rhythmic structures.
In addition to stylistically informed intentional accents the ODM should also emphasize and de-emphasize onsets just as human timekeepers do. While jazz drummers tend to have very well-developed and accurate ride cymbal patterns, these patterns must have a constant flux in accent and amplitude, each onset slightly different, some intentional and some simply human. Are there long-memory patterns in accent production as well as onset timing? How can this data be gathered and used in the ODM?
5) Maintain these predetermined patterns with less and similar accuracy compared to human timekeepers.
I imagine a process where human timing data is collected (either from previous studies or new ones) and algorithms are designed to emulate this data and trigger note onsets in the ODM. However, rather than working from clock time towards human time, I plan to approach the process in the same way a photographer puts an image in focus, spending time with pulse patterns that are too inaccurate (out of focus) to be considered musical, then making these patterns successively more human (focused) in accuracy. Metronomic time has become a very taken for granted standard of timing for onset events, especially in computer-based music, and initial goals include avoiding this kind of mechanical ISIP as the concept is developed.
Here are some things that the early generations of the ODM machine should NOT accomplish:
1) Interact with human performers or other programs.
Real-time interaction is absolutely a future goal of the ODM project. For the time being, however, my work will focus on the ODM’s rhythm generation software. While I hope that early generations of the ODM will be useful for live performances and recordings (a next step forward from my 2009 piece Descarga), these initial musical applications will not incorporate computer listening, onset detection, etc.
2) Generate new rhythms, compose, or improvise.
Algorithmic improvisation is another long term goal of the ODM project. At first the programming will be devoted to the problems associated with generating human-type ISIP rather than how to get a computer to compose convincing hierarchical rhythmic structures.
3) Communicate with other software platforms.
A rhythm computer able to play rhythms with this type of non-clock based algorithmic data would indeed be a new kind of sequencer. Integration of this kind of program into existing platforms (MIDI, DAW, etc) would require a fair amount of data translation. Good music software plays well with other music software. The ODM is intended to be a creative tool and should at some point be able to “plug in” to existing programs; for now this is not a priority.
And these are notably unknown:
1) Sample, synthesis, or hybrid sound realization?
It’s possible that the ODM program could simply send timing data to an existing drum machine. I hope that the timing program and sound synthesis will both take place in the Max/MSP/Jitter environment. All details are still to be determined.
2) Based on what “hard data” about human timekeeping?
Gathering highly specific information about rhythmic timing presents several difficult questions: What is a note onset? When can is be said to be “perceived;” at its exact beginning, when the player hears it, later when other musicians hear it, etc? Is the onset itself a germane piece of timing information or is a musical beat felt at some point before or after an attack? While there are many studies of ISIP among musicians and non-musicians, this data is often based on subjects’ computer key-clicking or drum pad input. Real musical timekeeping rarely uses either of these interfaces. Is there existing useful data measuring ISIP for musicians striking piano keys, cymbals, hand drums, plucking strings etc?
The ODM project may become a data gathering project at first–unlike some scientific studies my goal is to use data from musicians who are actively performing and parse this data using a certain amount of stylistic correlation–and this data might not yet exist.
3) Types of statistical, algorithmic, or random timing formulas.
Examination of available and new data on human musical ISIP and other periodic behavior should help determine the best modeling formulas. How different ways of crunching numbers affect the resulting patterns will be interesting. Many listeners, myself included, have become accustomed to hearing and enjoying various kinds of machine rhythm. Many theorists and composers have used statistical modeling in their musical work. The ODM will bring this kind of mathematics to bear on the sequencing of rhythm, and in the process my collaborators and I will have to learn to hear how fractal dynamics sound in various rhythmic cycles. I hope the ODM will be an aesthetically successful addition to the already well-developed field of rhythm sequencing.
This series of posts has examined various aspects of human timekeeping and musical examples that problematize the accepted notion that effective execution of rhythm in groove based music is essentially or ideally metronomic. My goal is neither to debunk the consensus idea of what “good time” is nor to decry the prevalence of mechanical rhythm. Rather, I hope to expand the functionality, nuance, and temporal palette of rhythm machines.
__________________________________
References
Benadon, Fernando. “Time Warps in Early Jazz.” Music Theory Spectrum, Vol. 31, No. 1 (Spring 2009), p. 1-25.
Benadon, Fernando. “A Circular Plot for Rhythm Visualization and Analysis.” Music Theory Online, Vol. 12, No. 3 (September 2007).
Chen, Yangqing et al. “Origin of Timing Errors in Human Sensorimotor Coordination.” Journal of Motor Behavior, Vol. 33, No. 1 (2001) pp.3-8.
Grahn, Jessica, and Matthew Brett. “Rhythm and Beat Perception in Motor Areas of the Brain.” Journal of Cognitive Neuroscience 19:5 (2007), pp. 893–906.
Hausdorff, Jeffery M. “Gait dynamics, fractals and falls: Finding meaning in the stride-to-stride fluctuations of human walking.” Human Movement Science 26 (2007) pp. 555–589.
Janata, Petr, and Scott T. Grafton. “Swinging in the Brain: Shared Neural Substrates for Behaviors Related to Sequencing and Music.” Nature Neuroscience Vol. 6, No. 7 (July 2003) pp. 682-687
Madison, Guy. “Fractal Modeling of Human Isochronous Serial Interval Production.” Biological Cybernetics 90, (2004), pp. 105–112.
Peng, C. -K., et al. “Quantifying Fractal Dynamics of Human Respiration: Age and Gender Effects.” Annals of Biomedical Engineering, Vol. 30 ( 2002), pp. 683–692.
Wright, Matt, and Edgar Berdahl. “A Survey of Computer Systems for Expressive Music Performance.” Final paper, Stanford Machine Learning (CS 229), Dec 2005.
