Finding Downbeats with a Relaxation Oscillator Douglas Eck Presented by Zhenyao Mo
Overview Purpose: to find downbeats in rhythmical patterns Method: using oscillators to model the behavior of rhythmical patterns, thus find downbeats
Problems in downbeats finding Imperfect timing –Motor noise –Deviations due to expressive timing
Why oscillators It has CYCLES It could respond to input perturbation, and finally align its cycles with some periodic components –Thinking of musical notes as input perturbation –Thinking of oscillator cycles as downbeats It can resist noise signals Perfect !!!
A Relaxation Oscillator Model dv/dt = -v(v- dv/dt = -v(v-θ)(v-1)-ω+Ω dω/dt = e(v-r ω) (v is voltage, ω is voltage recovery)
A Relaxation Oscillator Model dv/dt = -v(v- dv/dt = -v(v-θ)(v-1)-ω+Ω (input notes) dω/dt = e(v-r ω) (v is voltage, ω is voltage recovery)
Input note signals xxxx..xx.x.x… x is notes,. is a rest The base interval between notes is 125 time-steps x is encoded as 10000…0, one 1, . is encode as 00000…0, 125 0
Input note signals (continue) Two input signal strength –Low strength: ~0.08 (mean = ) –High strength: ~0.09 (mean = ) Three type of noise –No noise –-5%~+5% –-10%~+10% –(all apply to time/interval, not to signal strength)
Simulation 20 oscillators are used –Different initial conditions Run 8 pattern repetions, output 2 following pattern repetions, then recorded Results are compared to “P&E” Model
Discussion Why high strength input get worse results? –Oscillators are over-sensitive to “early” events –Later events may be more important for downbeat finding –The “strong” voltage in “early” events cause oscillators to “fire”, and unable to respond to subsequent events
Discussion (continue) When it fails –For those patterns where rests play an important role Future work –Using oscillator network (coupling and synchronizing) –Comparing different models