Gesture Signal Processing and Music
From CCRMA Wiki
Contents
Gesture and Measuring Movement
- What is a gesture? Physical. Bodily: hands, face, posture. Non-verbal. What do gestures communicate? Is a gesture more like a button or a handle?
- What is a musical gesture?
Techniques for measuring human movement
- Camera-based
- 2D image processing
- 3D Motion Capture using many cameras
- Goniometry: measuring joint angles with physically attached sensors
- Laetitia Sonami's Lady's Glove: http://www.youtube.com/watch?v=HYTrNOmSRSo&feature=player_embedded
- Electromyography: measures muscle activation by measuring electrical activity
- Pamela Z's controller (which I believe uses emg): http://www.pamelaz.com/VociMov.html
- With accelerometers attached to the body (e.g. wiimote, iphone)
Really basic physics
- Position: x
- Velocity: v = dx/dt
- Acceleration: a = dv/dt
- Proportional to force (Newton's 2nd law of motion: F = ma )
- Jerk: j = da/dt
Interlude #1: Open Sound Control
- Open Sound Control: http://opensoundcontrol.org/
- iPod Touch has 3 axis accelerometer
- We will use TouchOSC to transmit accelerometer data as OSC messages over wifi to Max/Pd
Interlude #2: Filtering of Audio Signals
- Can view any sound as being composed of sinusiodal waves at different frequencies
- A filter removes energy at only certain frequencies:
- A low-pass filter removes high freqs.
- A high-pass filter removes low freqs.
- A band-pass only lets certain middle frequencies through.
- Is differentiation a low-pass or high-pass filter? What about integration?
Back to movement and music
Accelerometers: How to analyze data
- How to distinguish orientation vs movement?
- Can get jerk via difference (as approximation of differentiation)
- How to get velocity from acceleration (or position from velocity)?
- True integrator (will eventually overflow):
y = y_prev + x.
- Leaky integrator as approximation of integration:
y = a*y_prev + (1-a)*x
- True integrator (will eventually overflow):
- Thresholding - detecting specific events
- max: past. pd: threshold
- Classification through training of Machine Learning algorithms
- e.g. Rebecca Fiebrink's Wekinator: http://wekinator.cs.princeton.edu/
Mapping to Sound
- How do we create a musically expressive instrument from sensor data?
- We need to map the data from one domain to another appropriately:
- linear: to map [a b] to [c d] use
z = (s-a)*(d-c)/(b-a) + c
- exponential: more appropriate in many situations since perceptual qualities are often logarthmic.
- any arbitrary function
- linear: to map [a b] to [c d] use
- max: scale, pd: ???
- How do we decide the best mapping? (i.e. How do we select what physical parameter controls which musical parameter?)
- this is an art, not a science!
