• Abstract and Figures
• Public Full-text

Techniques to Foster Drum Machine Expressivity Je A. Bilmes <[email protected]> Perceptual Computing Group MIT Media Lab oratory Massachusetts Institute of Technology Cambridge, MA 02139 Abstract Electronic drum machines need algorithms to help them pro duce \expressive- sounding" rhythmic phrases. In [Bilmes, 1992], I claim that three p erceptually separate ele- ments characterize p ercussiverhythm: metric content, temp o variation, and deviations (for- merly called event-shifts). Herein, I demonstrate that algorithms based on this mo del may considerably facilitate repro duction of expressiverhythm. I describ e one such algorithm which extracts the separate elements from a p ercussive p erformance. The p erformance is then resynthesized with varying degrees of temp o variation and deviations. Without the de- viations, the p erformance sounds mechanical. With them it sounds rich and alive. Conse- quently, I claim that we should b egin a concentrated study on the separate elements char- acterizing p ercussiverhythm, particularly deviations. To this e ect, development has b e- gun on a graphical drum machine program with which deviations may b e explored. 2 Rhythm 1 Intro duction There is no doubt that music devoid of b oth har- To err is human. Yet most users of drum machines monyandmelodymay still contain considerable and music sequencers strive to eliminate \errs" expression. Percussivemusic is a case in p oint, in musical p erformance. In fact, some computer 1 as anyone who has truly enjoyed traditional mu- musicians never turn o the quantize option, de- sic from Africa, India, or Central or South Amer- stroying forever \ aws that make the p erformance ica knows. sound sloppy." At the same time, other computer Unsuitable for p ercussivemusic however, musicians complain ab out the mechanical quality previous representations of expressive timing of computer music. They call for the development ([Clines, 1977], [Ja e, 1985], [Schloss, 1985], of techniques whichwould enable computers to [Repp, 1990], [Wessel et al., 1991], [Anderson and sound b etter, i.e. more \human." Kuivila, 1991], [Anderson and Bilmes, 1991], and There are two orthogonal criteria of p er- [Desain and Honing, 1992]) can all (with the ex- formance. The rst is sheer technical pro ciency. ception of [Desain and Honing, 1992]) b e reduced Clearly, computers have long surpassed humans to temp o variation. on this axis. The other is expressivity, something In [Bilmes, 1992], I intro duce a new mo del more elusive, something that gives music its emo- of rhythmic expressivity. Sp eci cally, I state that tion, its feeling, its joy and sorrow, and its human- b eat-based rhythm can b e characterized bythree ity. Music exudes humanity; computer music ex- comp onents: metric structure, temp o variation, udes uniformity. This, I strive to eliminate. and deviations (formerly called event-shift mo d- Author's current address: Computer Science Division, els). In this pap er, I argue that deviations are EECS Department, U.C. Berkeley, Berkeley, CA 94720. most imp ortant for p ercussive and non-Western <[email protected]> 1 I use \computer musician" to refer to anyone who uses music, and that they are indisp ensable study for a computer to create music, and \computer music" to re- any drum machine architect wishing to create an fer to music created thereof. expressive sounding pro duct. When we listen to or p erform music, weof- ten p erceive a high frequency pulse, frequently a binary, trinary, or quaternary sub division of the 2 musical tactus . What do es it mean to p erceive 3 tatum . this pulse, or, as I will call it, 2000 . erceiving the tatum do es not necessarily P . king in the mind, likea clock. imply a conscious tic . 354 Often, it is an unconscious and intuitive pulse Tatums/Minute that can b e broughtinto the foreground of one's 12345678 twhenneeded.Perceiving the tatum im- though Tatum Num (A) plies that the listener or p erformer is judging and ticipating musical events with resp ect to a high an +15% frequency pulse. It is a natural p erception, p er- 0% haps similar to the illusory contour in the well -15% known picture on the frontcover of Marr's b o ok Deviation: % of current tatum duration. [Marr, 1982]. 12345678 ays explicitly stated in The tatum is not alw Tatum Num (B) usic. How, then, is it implied? The a piece of m Tempo = 300 tatums/minute. tatum is the lowest level of the metric musical hi- erarchy. Often, it is de ned by the smallest time Figure 1: EquivalentTemp o Variation and Devi- interval b etween successive notes in a rhythmic ation Representations phrase. For example, two sixteenth notes followed by eighth notes would probably create a sixteenth note tatum. Other times, however, the tatum is memb ers of the ensemble slightly adjusting their not as apparent; then, it might b est b e describ ed own p ersonalized temp o. Furthermore, the temp o as that time division which most highly coincides change needed to represent deviations in a typi- with all note onsets. cal p erformance would b e at an unnaturally high The tatum provides a useful means of de n- frequency and high amplitude. Imagine, at each ing twocomponents of b eat-based rhythm, temp o successive tatum, varying the temp o b etween 2000 variation and deviations. Tatums pass byata and 354 tatums p er minute (Figure 1A). Perceptu- certain rate, and may b e measured in tatums p er ally, this seems quite imp ossible. However it seems minute. Temp o variation may b e expressed as quite reasonable to assume that a p erson could, at tatum duration (in seconds) as a function of tatum a constant temp o of 300 tatums p er minute, play numb er. Similarly, deviations may b e expressed every other note 15 p ercent of a tatum early (Fig- as deviation (in seconds) as a function of tatum ure 1B). In other words, although they mightbe numb er. That is, a deviation function determines mathematically equivalent, temp o variation and the amountoftimethatanevent metrically falling deviations are di erent in more imp ortantways { on a particular tatum should b e shifted when p er- they are distinct functionally and conceptually. formed. Note that temp o variation is de ned p er The previous paragraph suggests that there ensemble, whereas deviations are de ned p er p er- must b e some (p er p erson) upp er limit on temp o former. oscillation frequency. That is, any p erformance A common question asked is, if we assume variation not accounted for by temp o variation b e- temp o variation is also p er p erson, why not just cause of its high frequency must b e owing to \de- use temp o variation to represent deviations in a viations." The following algorithm was develop ed p erformance? That is, is there mathematically with this assumption in mind. any di erence b etween temp o variation and devia- tions? The answer is no, there is no mathematical di erence. Either can represent p erformance tim- 3 Timing Extraction Algo- ing. There is, however, a p erceptual di erence. When listening to or playing in a drum or rithm jazz ensemble, there are times when the temp o In [Bilmes, 1992], I p oint out the necessityofan is considered constant, even though memb ers are algorithm that extracts deviations from a p erfor- playing o the b eat. Notice, the concept of b e- mance. Presented herein is one that extracts the ing \o the b eat" suggests that there is deviation quantized score, the temp o variation, and the de- from some temp o generally followed by the ensem- viations. The input to the algorithm is a list of ble. There is no concept, however, of individual attack times. 2 The stroke of the hand or baton in conducting, or, of- The algorithm is given complete metric ten, the main quarter note b eat 3 knowledge of the p erformer. That is, it knows the When I asked Barry Verco e if this concept had a term, he felicitously replied \Not until now. Call it temporal time signature, the numb er of tatums p er b eat, atom,ortatom." So, in honor of Art Tatum, whose tatum the numb er of b eats p er measure, and where the was faster than all others, I chose the word tatum. in time, the reference instrumentinformstheen- b eginning of the measure is (the answer to \where semble what the temp o is. The p erformance in- is one?"). There are twoversions of the algorithm; strument, dep ending on whether it is dominant one is primarily for p ercussivemusic, the other is in the ensemble (such as a lead drum), controls slightly more general. when the temp o sp eeds up and slows down. That Version I of the algorithm requires a rep et- is, wesay the reference instrument de nes the itive reference instrument (such as a b ell, a clave, temp o, and the p erformance instrument controls or a bass). The reference instrumentisusedto the temp o. Therefore, when obtaining the timing extract temp o and must rep eatedly play a known of a dominant p erformance instrument, welook pattern. The pattern p erio d mustbeaninteger slightly ahead, and compute multiple of the measure duration. Percussivemu- sic normally contains such an instrument, so this n+C X 1 is not an unreasonable requirement. The algo- 0 T [n]= T [k ]; rithm pro duces the expressive timing of a perfor- C +1 k =n mance instrument relative to the reference instru- where C is a parameter determining howfarinto ment.

Load more

  8

Copy Link