Brian Gurewitz
Final Project – MUS 220A
Instructors: Chris Chafe, Romain Michon, Holly Herndon
Sound of Sleep
Binaural Audio File:
4-channel Audio Files:
Binaural Chuck Code:
Sound Generators (using DataReader algorithm):
eeg_fpzcz_sleep_Player.ck – sin wave generator that reads in front EEG data to modulate pitch
eeg_pzoz_sleep_Player.ck – sin wave generator that reads in rear EEG data to modulate pitch
eog_sleep_Player.ck – generates music using the Rhodey STK instrument. Notes selected at random from the first six notes of a major scale. Rhythms selected at random to be either 2 seconds, 1 second, or 0.5 seconds. This is a modified version of an example file found in the Chuck documentation. Horizontal ocular motion is read in to pan the sound from left to right in accordance with eye movement.
emg_sleep_Player.ck – noise generator that reads in emg data to mimic the effect of breathing
Source Data:
PhysioNet Lightwave (link) --> Database: Sleep EDF --> Patient ST7022J0 (Female, 35 years, recorded in 1994)
Background and Project Description:
The general idea behind this project was to create, in 4-channel, a spatial sonification of sleep using real biological measurements taken while a subject was sleeping.
Sleep is a time when the brain undergoes multiple cycles of coordinated electrical activity. Sleep is divided into stages depending primarily on the kinds of waves observed from the EEG. During Stage 1, alpha waves predominate (8-13 hz). During Stage 2, theta waves take over at around 12-14 hz. These patterns are known as sleep spindles. Stages 3 and 4 are considered deep sleep, and they are characterized by the presence of low-frequency (0.5-2 hz), high-amplitude delta waves. Finally, REM sleep is most similar to the waking state in terms of electrical activity and is characterized by rapid eye movements (REM). REM sleep is the stage where dreaming occurs.
The patient chosen was a 35 year old female who had a sleep recording done in 1994. This recording is now available for free via PhysioNet's Sleep-EDF database. The patient had two EEG measurements taken: first at the fpz and cz locations (forehead and top center), second at the pz and oz locations (mid-back and back of head). I recreated these two data streams using sin oscillators with slightly differing tessitura. I split them in the 4 channel space according to where on the head they originated.
The patient also had horizontal EOG measurements and EMG measurements taken. The horizontal EOG is a way of tracking eye position from left to right. The electrodes are set up such that when the eyeball moves towards the right, a positive potential is incurred (and negative for left). I used this data as panning information to be able to send sound to the right or left in accordance with the patient's eye motion. EMG measured muscle activity below the head, which I've interpreted to mean breathing activity given the steady rise and fall observed in the data. I used a Noise generator to emulate the rise and fall of the breath, and I exponentiated the data points in order to create a differentiation high enough for the listener to perceive the breath.
Results:
The final result in the audio files above consists of a segment of Stage 2 sleep (0:00 – 2:30) and then a segment of REM sleep (2:30 – 7:20). Due to the relatively fine distinctions between the sound of Stage 2 and Stage 3-4 (Alpha waves don't sound that different from Delta waves within my overall model), I felt it was better to show the contrast between non-REM sleep and REM sleep.
Over stage 2 sleep I added a piano composition that does not read in real sleep data. The goal was to emulate the serenity and the mystique of falling into sleep. For the REM sleep I used an algorithm to generate random music based on a series of notes in the major scale. Since REM is characterized by lots of eye motion, I used the EOG data to pan the random music back and forth between the eyes, intending to demonstrate the exploratory and associative quality of dreams. You can also hear how the EEG measurements during the REM stage are more variable than during sleep stage 2, more like what you would find in an awake patient.