Two layers run in parallel: (1) continuous raw streaming for immediate “voice” motion, and (2) a slower computed feature layer that is translated into macro-scale sonic / spatial behaviors.

Data source (e.g. CSV) → streamer → OSC → Ableton Live (M4L) → audio / spatial / haptic output

Polysomnographic recordings (EEG + EOG/ECG/EMG) are streamed from CSV (typically using original 512 Hz sampling rate) and routed as individual control streams. In the music software, these streams can (a) continuously modulate synthesized tones (think a choir of continuously warbling voices), and/or (b) be downsampled/quantized to musical subdivisions (as orchestral instruments).

Raw layer: each channel as a continuous voice or (downsampled) instrument.
Feature layer: band-power (calculated across different selected eeg electrodes) generate larger contours and event-like cues (slower updates) that can steer tonality (key/scale), timbre, diffusion, and stability. In the current prototype, a custom M4L global transport shifter compares delta/theta/alpha/beta, chooses the dominant band, maps it to one of four preselected global scales, and derives root from the streamed /eeg/segment_index by an editable fixed semitone step (currently +7 for perceptually clearer stage-to-stage motion).

The interface is designed for exploratory listening—like moving a microscope around an inner landscape.

• Solo / mute / rebalance channels to compare regions (e.g., frontal vs occipital; left vs right).
• Contrast brain vs physiology by emphasizing EEG or isolating EOG / ECG / EMG.
• Stage comparison (wake → N1 → N2 → N3 → REM) by switching excerpts or selecting portions of recordings.
• Focus listening modes (e.g., “micro-detail” with or without broader “sleep weather” patterns)

The system can compute band power of different wavelengths (alpha, beta, delta, etc.) and map each to perceptual qualities such as density, agitation, brightness, stability, depth, and spatial diffusion.

Features are computed on short windows (seconds) with overlapping hops, then streamed (via OSC) in sync with (but at a slower update rate than) the raw layer.

The band-to-scale mapping was chosen by balancing modal brightness with tonal ambiguity and affective character. Delta is currently mapped to Iwato because of its sparse, hollow, subterranean quality, which evokes deep sleep better than a conventionally dark mode alone. Theta is mapped to Whole Tone because theta is treated as dream-state material: floating, liminal, and weakly grounded, which Whole Tone expresses through symmetry and tonal suspension. Alpha is mapped to Major because it best represents calm, stable, coherent wakefulness. Beta is mapped to Lydian Augmented because it conveys the highest degree of brightness, lift, shimmer, and activation. The result is a four-state musical logic of deep sleep → dreaming → relaxed coherence → heightened cognition.

In the current transport-scale shifter, delta/theta/alpha/beta compete as a dominant-band selector feeding a four-scale whitelist, so the exact scale assignments remain editable in the Max patch. Sigma currently remains available as a separate spindle/shimmer control layer rather than acting as the main scale selector.

Sleep stage / segment identity also shapes tonal structure. The streamed /eeg/segment_index is converted to global root by stepping through an editable fixed semitone interval; the current setting moves each successive segment by 7 semitones (ascending fifths). This was chosen because smaller root shifts felt too subtle relative to the density of simultaneous sonic information: fifths read more continuously and noticeably, while still leaving the interval under-the-hood editable.

(Mapping is intentionally adjustable: the same features can drive sound, light, spatialization, or haptics, depending on installation context.)

Auxiliary physiology is treated as an equal musical / physical partner.

• ECG (raw + derived): the raw ECG channel streams continuously, and a parallel detector extracts discrete heartbeat events. The system emits a short /eeg/ecg/beat pulse at each detected beat, plus continuous descriptors /eeg/ecg/ibi_ms (inter-beat interval, held between beats) and /eeg/ecg/bpm (computed as 60,000 / IBI, also held between beats).
• Heartbeat percussion (event-timed, not tempo-quantized): beat pulses trigger a drum voice in Ableton. A small Max “edge + refractory” stage prevents re-triggering: it fires only on the rising edge of each beat pulse and enforces a minimum interval (debounce / refractory) so the drum does not double-hit when upstream pulses are sampled.
• Heart-rate modulation (IBI / BPM): IBI and BPM are used as slow control signals to shape the character of the ECG drum chain (e.g., by modulating the gain of selected layers and/or mix parameters), so changes in cardiovascular state register as changes in intensity and balance rather than as extra triggers.
• ECG “energy envelope” (optional layer): the continuous ECG waveform can also be treated as an envelope (rectified + smoothed) to gently animate the gain/presence of the ECG chain and/or the broader physiology bus (ECG/EOG/EMG), adding subtle “organismic” motion beyond discrete beat hits.
• EOG + EMG: eye (EOG) and muscle (EMG) channels stream as continuous signals and are rendered as their own instrumental voices (synth / sampled textures). The mappings are intentionally chosen to echo the rhythmicity and complexity of each signal.

Reliability note: where needed, simple cleanup is applied at the trigger stage (edge detection + refractory/debounce) to ensure one clean event per beat, while keeping raw streams available for inspection and alternative mappings.

These options are chosen before streaming (in the CSV→OSC streamer and run menu). Real-time controls (mute/level/mapping depth/etc.) live in the Ableton/M4L layer.

Configured: Menu prompt = asked each run; Menu auto = set silently.

Streamer (CSV → OSC)

Additional developer-level parameters exist (naming, network robustness, detector tuning), but are not exposed in the run menu.

In the current Ableton/M4L layer, the streamed segment index is converted to global root using an editable fixed-step formula (currently 7 semitones), while the dominant band among delta/theta/alpha/beta selects one of four editable global transport scales.

Run menu (batch playback)

• Select multiple staged excerpts and play them sequentially.
• Optionally precompute feature caches (and exports) first to keep playback gapless.
• Choose aggregation presets for features (event/rem/custom) and optional boosts.
• Optional per-stage exports: reports (text/plots) and video renders (mp4/gif) in a dedicated folder structure.

A set of listening cues: these are common tendencies, not guarantees. Sleep stages are variable, but they have detectable—and perceptible—signatures.

• Wake: often more alpha/beta presence; faster “edge” motion.
• N1 (drowsy onset): alpha tends to soften; theta often emerges as drifting depth.
• N2: spindles can appear as brief shimmering bursts; K-complexes can read as sudden large-scale gestures.
• N3 / SWS: delta dominance—slow, heavy waves; strong synchronizations can feel like collective breathing.
• REM: mixed-frequency textures; EOG gestures may become more prominent.

Orchestral instrument mappings: downsampled / quantized streams drive MIDI synth and sample instruments at musically meaningful rates. This supports coherent instrumental texture.

Synth choir mappings: each channel continuously modulates a dedicated synth voice in real time. This preserves micro-timing and yields a richly textured “choral” surface where rhythms and synchronizations become audible through fine pitch motion.

In each case, band power features can be computed and added in as perceivable layers (of tonality, timbre etc.)

These approaches can be mixed (e.g., continuous choir + quantized orchestra layer), and the same features can drive sound, light, spatialization, or haptics depending on installation context.

The physiology layers can also be mapped direct to synth choir, but it seems more fitting that these —or at least the ECG streams— remain mapped to percussive and synth sounds, as (the ECG in particular) has been carefully calibrated (i.e. with a custom-built MAX patch to filter most likely peak values that correspond to heart beats).

EEG labels follow the International 10–20 naming convention (F/C/T/P/O, left/right, midline).

(Exact channel availability depends on the recording montage and the excerpt.)