Brain wave music for public science events: a 3-step setup

The cortex hums. Across the skull, populations of neurons oscillate at frequencies anyone who has closed their eyes in a quiet room has felt: Delta below 4 hertz in deep sleep, Theta from 4 to 8 hertz in drowsy drift, Alpha from 8 to 12 hertz when the eyes close and attention softens, Beta from 13 to 30 hertz during active engagement and problem-solving.

For most of electroencephalography’s century-long history, those oscillations lived as wavy ink on paper or as a slow band of colour scrolling across a hospital monitor. At a public science event, the audience walks past a screen, looks for two seconds, and moves on. Brain wave music exists because the same signal can be routed somewhere else: into a speaker array, where a room full of visitors hears attention arrive, settle and drift in real time.

The pipeline is not magic. It is a constrained, three-stage workflow — acquire, decompose, map — built from ordinary signal-processing components and a small set of musical decisions. Done carefully, it produces a brain wave music setup for public events that is audibly legible, ethically honest and resilient to the hum of a crowded exhibition hall.

Why sonification works where a chart does not

Public engagement with neuroscience has long leaned on the visual: a printed spectrogram, an animated brain on a touchscreen, a slow colour sweep across a head map. Those representations can be precise, but they require the visitor to stand still, decode a legend and trust a key. Sonification routes around some of that friction. A non-specialist audience already knows how to listen.

When Alpha power rises in occipital channels after a participant closes their eyes, the installation can respond with a softer chord, a slower texture or a more sustained tone. The audience does not need to understand every line on the spectrogram to notice that the sound has changed. The visual display still matters: it anchors the experience and gives the demonstrator something concrete to point to. But the sound provides an immediate sense of movement, even for visitors who arrive halfway through the session.

That immediacy should not be confused with transparency. Sound is a translation, not a direct broadcast from the cortex. The designer has already selected the channels, filtered the signal, calculated band power, normalized it against a baseline and decided which musical parameters will respond. The audience hears a designed interpretation of neural activity. That is not a weakness. It becomes a problem only when the installation presents the interpretation as if it were an unedited inner soundtrack.

The lineage is older than most visitors assume. In the early history of EEG, researchers explored ways to make electrical brain activity audible rather than merely visible. Later, composers including Alvin Lucier used Alpha rhythms to influence percussion and electronic sound in performance. The contemporary public-engagement version inherits that experimental logic but works inside tighter constraints: a museum foyer, a short session, a changing queue and an audience that came to see an installation, not to read a methods section.

A brainwave installation is a translation problem, not a recording. What the audience hears should be a faithful simplification of the signal, never a literal dump of every oscillation the headset can see.

The three steps, in order

A reliable brain wave music setup for public events runs through three stages in sequence. Skipping one — mapping raw signal straight to a synthesizer, for instance — usually produces unstable, muddy sound and an unconvincing demonstration. Each stage has its own failure modes, and each rewards modest, well-understood choices over ambitious ones.

Step 1 — Acquire clean EEG in a noisy room

Public venues are hostile to electrophysiology. HVAC systems, lighting equipment, laptops, wireless transmitters and long cable runs all compete with a signal measured in microvolts. The visitor is another source of movement: they blink, speak, turn their head, clench their jaw and adjust the headset. The first task is therefore not to find the most impressive musical mapping. It is to choose hardware and a layout that can survive the room.

The electrode type needs to be described accurately, because the words are not interchangeable.

• Dry electrodes make contact with the skin without gel or saline. They are quick to fit and well suited to short public sessions, but contact quality can be sensitive to hair, pressure and movement.
• Wet electrodes use conductive gel or paste. They generally provide a more stable electrical contact, but preparation takes longer and the participant may need more help before and after the session.
• Semi-dry electrodes use a small amount of saline or another conductive medium, often held in a sponge, reservoir or polymer contact. They sit between the two approaches: faster and cleaner to use than a conventional gel cap, but not genuinely dry.

That distinction matters in both the technical setup and the public explanation. A flexible consumer headset with saline-soaked sponges or saline-loaded polymer contacts is semi-dry, not dry. A headset with polymer contacts that touch the skin without any conductive liquid is dry. The two designs may look similar on a table, but their preparation, maintenance and failure modes are different.

Consumer EEG headsets are often attractive for outreach because they can be fitted quickly and do not require a dedicated preparation room. More demanding installations may use a wet-electrode cap or a research-grade system with more channels and stronger control over impedance and signal quality. The trade-off is operational rather than purely scientific. More channels and better contact can improve the data, while also increasing setup time, cleaning requirements and the amount of training needed from the demonstrator.

The installation should make that trade-off visible instead of hiding it. If visitors see a participant being prepared for several minutes, that is an opportunity to explain why electrodes need a reliable connection to the scalp. If the event uses a quick-fitting headset, the explanation should include its limitations: movement and contact changes can affect the signal, and a short public demonstration is not equivalent to a clinical EEG recording.

Positioning helps before any software is opened. Keep the participant away from obvious sources of electromagnetic and mechanical noise where the venue allows it. Avoid placing the chair directly beside a dimmer rack, power distribution unit or large motor. Use the shortest practical cable run, secure loose leads and keep the laptop and audio interface arranged so that cables do not cross the participant’s body unnecessarily.

A dark or visually calm background can also reduce movement. Participants tend to turn their heads when they are watching a crowded screen or responding to people behind them. A simple instruction to keep the jaw relaxed and the eyes open or closed for a defined part of the session is more useful than a vague request to remain still.

The demonstrator should know the channel layout in plain language. Frontal, central and occipital regions are more useful to a public audience than a string of electrode labels, provided the explanation does not imply that one electrode corresponds to one thought or one mental state. That vocabulary returns in the musical mapping stage and lets the audience understand what the installation is actually using.

Step 2 — Filter the signal and decompose it into bands

Raw EEG is a thicket. Eye blinks produce sharp transients that can be much larger than the cortical activity beneath them. Jaw clenching and forehead tension inject muscle noise across a broad frequency range, including frequencies that may overlap the bands used for the installation. Head movement changes electrode contact and can create slow drifts or abrupt jumps.

Before any music can be made, those artefacts have to be identified and controlled.

A practical processing chain often includes:

• A band-pass filter that keeps the frequencies relevant to the demonstration and suppresses slow baseline drift and much of the high-frequency muscle noise. A range around 1 to 40 hertz can be a reasonable starting point, but the exact settings depend on the headset, the sampling rate and the intended mapping.
• A mains-frequency strategy suited to the local electrical grid. In regions using 50 hertz power, interference commonly appears around 50 hertz; in regions using 60 hertz power, it commonly appears around 60 hertz. Those frequencies are outside the conventional 13–30 hertz Beta range, so mains interference does not sit exactly on top of low Beta. It can still contaminate the recording through poor grounding, cable coupling, amplifier behaviour or harmonics, and it may become audible if the processing chain is too broad. A notch filter or another suppression method can be used when the interference is present, but it should not be described as a Beta-band problem.
• An artefact-removal step, which may involve channel inspection, regression, independent component analysis or simpler rejection rules. Independent component analysis can help separate ocular and muscular sources, but it is not an automatic guarantee that every rejected component was noise. In a short public session, a transparent and conservative method is often preferable to an elaborate pipeline that the operator cannot explain.
• A decomposition into the frequency bands used by the installation: Delta below 4 hertz, Theta from 4 to 8, Alpha from 8 to 12 and Beta from 13 to 30. Band power can be estimated with a sliding-window fast Fourier transform, a wavelet transform or another spectral method.
• A normalization step that expresses each band’s power relative to that participant’s own baseline. A short resting period at the start of the session is usually more informative than comparing one visitor with another using an absolute threshold.

The baseline is the unglamorous step that makes the system feel responsive. Without it, one participant’s higher Alpha power may be treated as if it were the same event as another participant’s higher Alpha power, even though headset placement, scalp contact, hair, movement and individual physiology all affect the measurement. The music may then flatline for one visitor and become overactive for the next.

Normalization does not solve every interpretation problem. It only makes the mapping more individual and more stable. A rise in a band relative to baseline is still not a simple label for an emotion. Alpha power can vary with eye closure, visual input, attention and other factors. Beta-range activity can be affected by muscle tension. The demonstrator should be able to explain these complications without turning a three-minute exhibit into a lecture on every limitation of EEG.

Responsiveness also depends on the time window. A very short window reacts quickly but gives an unstable estimate of power. A longer window produces smoother control but makes the music feel delayed. Public installations usually need a compromise: enough averaging to prevent every blink from becoming a dramatic musical event, but not so much smoothing that the visitor cannot hear a connection between what they do and what changes in the room.

Step 3 — Map bands to musical parameters with discipline

Raw brainwaves do not naturally form harmonious melodies. They must be constrained. This is the step that determines whether the installation sounds like a meditation app, an alarm system or a working scientific instrument.

A practical mapping assigns each frequency band to a limited set of sonic roles. The table below shows one workable default. Curators can revise it, but the principle of one band, one instrument family and one expressive range keeps the result intelligible.

These labels need careful wording. They are useful descriptions for a sonification, not diagnostic categories. A participant does not become relaxed because the system detects Alpha, and a brighter sound does not prove that they are concentrating. The installation is choosing a way to make changes in the signal perceptible.

Three further rules keep the mapping honest.

First, constrain the output to a single scale or a small harmonic vocabulary: a pentatonic scale, a modal collection or a limited major-triad cluster. The point is not to make the music bland. It is to ensure that small fluctuations remain listenable. A public audience should hear a change in texture or harmony rather than be punished by a random sequence of dissonant notes.

Second, map band power to musical parameters along smooth curves rather than abrupt thresholds. A small change in measured power should produce a small change in the sound. If a single threshold flips a pad from silent to loud, the audience will hear the arbitrary rule rather than the neural data. Continuous mapping is usually more convincing, especially when combined with a bounded output range.

Third, smooth the control signal. A single blink should not trigger a percussion hit, and a brief electrode movement should not send the entire system into a new key. Exponential smoothing, moving averages or median filters can reduce these jumps. The settings should be tested with deliberate blinks, jaw clenches and head movements so the operator knows how the installation fails before the audience finds out.

It is also useful to separate slow and fast musical decisions. Slow changes in band power can control harmony, timbre or spatial position. Faster fluctuations can influence note density or modulation depth, but they should be limited. Trying to turn every sample or every short-term fluctuation into an event creates noise, not responsiveness.

Spatial audio is one of the underused advantages of EEG sonification. If the system has channels associated with frontal, central and occipital regions, the corresponding sound groups can be arranged across a similar front-to-back layout. This does not mean that the installation is literally projecting thoughts from different parts of the skull. It gives the audience a spatial metaphor for the channel arrangement and makes the room itself part of the explanation.

A small screen showing the live spectrogram or the changing band values can reinforce that metaphor. The visual layer should not duplicate every technical detail. It needs to show enough for the visitor to connect a visible change with an audible one. The best installations let the audience move between three levels of understanding: something changed in the participant, the graph registered a change, and the sound made that change perceptible.

Practical details for a one-day exhibit

A workable brain wave music exhibit can be staged with a small team, one table and a clear run-of-show. The numbers below are starting points rather than universal requirements. The venue, headset and level of scientific ambition will determine the final design.

Hardware and signal path

The basic hardware usually includes an EEG headset or cap, a laptop, an audio interface and a speaker system. A four- to six-speaker arrangement can provide more spatial information than a single pair of monitors, but even a modest stereo setup can support a strong demonstration if the mapping is restrained.

Keep the EEG acquisition path and the audio path easy to inspect. Label the main cables, mark the headset’s channel orientation and keep a spare set of contact materials ready. If the system depends on wireless transmission, test it in the actual room rather than assuming that a connection that works in the laboratory will remain stable beside dozens of phones and laptops.

A live spectrogram is useful, but it should not become the main attraction. Visitors came to hear the translation, not to watch a technician diagnose a loose electrode for five minutes. Give the operator a clear signal-quality view and give the audience a simplified display.

Software choices

The acquisition layer may use OpenBCI, BrainFlow or a manufacturer’s software development kit, depending on the hardware. MNE-Python and EEGLAB are common tools for filtering, inspection and decomposition. Pure Data and custom scripts offer open and flexible routes into MIDI, synthesis and spatial routing. Max/MSP is also widely used in interactive sound work, but it is proprietary rather than open source. The same qualification applies to many vendor SDKs: they may be well documented or freely available for a particular device, but that does not make the entire stack open source.

A more accurate description is that a public EEG sonification system can be assembled from a mixture of open-source tools, proprietary applications and device-specific software. That mixed ecosystem is not a problem. It simply needs to be described honestly, especially in a research laboratory or public institution where visitors may reasonably ask how the installation was built.

The processing pipeline should be tested end to end with recorded data as well as with a live participant. A prerecorded signal makes it possible to check the musical mapping, speaker levels and visual display without depending on the next person’s electrode contact. A live test then reveals the practical problems: wireless dropouts, delayed sound, noisy channels and the time required to reset the system.

Environment and staffing

A quiet corner of the venue is worth more than an elaborate software patch. Give the participant a stable chair, enough room for the demonstrator to fit and remove the headset, and a clear boundary that keeps passing visitors from brushing against cables. Lighting should make the participant comfortable without placing a large, noisy display directly in their line of sight.

A two-person team is useful. One person manages fitting, baseline recording and signal quality. The other explains the installation to the participant and the surrounding audience. Asking the participant to interpret their own signal while also holding still and listening can overload a short session. The demonstrator should carry most of the explanation.

A simple session might include a baseline, headset fitting, a short eyes-open period, a short eyes-closed period and a final explanation of what changed. The exact timing can vary. What matters is that the audience understands which part of the experience was intentional and which changes may have come from movement or contact.

Before opening, test the system in the order the visitor will encounter it:

1. Fit the headset on a team member and confirm that the channels behave plausibly.

2. Record a baseline and check that the band-power estimates do not drift uncontrollably.

3. Deliberately introduce a blink, a jaw clench and a small head movement to see how the pipeline responds.

4. Check that the smoothing prevents abrupt musical jumps without making the system feel inert.

5. Confirm that the speaker layout, display and explanatory text all refer to the same channels and bands.

6. Run a complete session from consent to headset removal, including cleaning or replacing the contacts.

The installation should also have a fallback mode. If the signal becomes unusable, a quiet explanatory state or a prerecorded demonstration is better than pretending that the music still represents the participant. The audience will forgive a technical interruption. They will not benefit from a fabricated interpretation.

What a curator should say out loud

Public trust is the asset. A single overstatement — that a headset reads the visitor’s mind, reveals their personality or diagnoses concentration — can damage the credibility of the whole event. The boundaries should be stated before the music begins, not hidden in small print.

The headset is not a medical device and the installation is not a diagnostic examination. The music is a constrained translation, not a readout of emotion or personality. The signal has been filtered and selected; visitors are not hearing every electrical event recorded by the electrodes. One participant’s data is not a general model of the brain, and a musical change does not prove that a particular mental state has been detected.

Consent and data handling belong in the same conversation. Visitors should know whether their signal is stored, whether it leaves the laptop, how long it will be retained and whether it will be used for anything beyond the event. A short public demonstration does not need a complicated legal performance, but it does need clear language. If the data is discarded after the session, say so. If it is retained for research or evaluation, explain that before fitting the headset.

The most convincing explanation is often the least theatrical one: the system measures electrical changes at the scalp, extracts selected frequency ranges, and uses those values to control parameters in a sound design. That process is already rich enough. There is no need to add claims about hidden thoughts.

Pick the scale to match the room, the band-to-instrument pairing to match the audience, and the filter to match the science. The cortex will hum either way.

Before the doors open

The final preparation is less glamorous than the mapping, but public exhibits usually fail at the edges. Batteries run low, replacement contacts disappear, a speaker points at a hard wall and a carefully prepared explanation is forgotten under the pressure of a queue.

Keep the headset battery charged and have a backup plan for the acquisition connection. Prepare saline sponges, semi-dry reservoirs, gel or replacement contacts according to the actual electrode type in use. Do not label a saline contact as a dry electrode simply because it does not use conventional gel.

Restart the laptop and audio interface before the event, then test the complete chain rather than checking each application in isolation. A signal visible in the acquisition software is not enough. It must also survive filtering, band-power estimation, normalization, smoothing, synthesis and speaker routing.

Use a fresh baseline for each participant. The baseline is part of the demonstration, not dead time: it gives the audience a reason for the first quiet interval and helps the system avoid treating every visitor as if they had identical signal characteristics.

Set the musical scale in advance, load a limited sound palette and check the levels from the place where visitors will stand. Match the speaker layout to the explanation of the electrode layout. Keep the participant information visible, including the non-medical disclaimer and the data-retention policy.

Most importantly, rehearse the explanation until it sounds like a normal sentence rather than an apology. The installation is not pretending to read the mind. It is showing how a measured signal can be filtered, summarized and translated into sound — and how much interpretation enters that process along the way.

Closing principle

Brain wave music sits at a productive crossroads. It takes a private electrophysiological signal and renders it, in real time, into a room-scale phenomenon that anyone with functioning ears can follow. Done well, it can teach a non-specialist audience more about neural oscillation in a few minutes than an hour of slides, because the visitor is not only watching a brain-shaped animation. They are hearing a signal change in response to a participant’s actions.

The workflow remains simple: acquire, decompose, map. The choices inside each step are where the curator’s judgement shows. Choose electrode terminology that matches the hardware. Suppress mains interference at the frequency where it actually occurs rather than assigning it to Beta. Use a software stack that is powerful without pretending every component is open source. Build enough smoothing and fallback into the system that the music can remain expressive without claiming more certainty than the data supports.

The visitor who leaves understanding that closing the eyes can change the signal, that the signal must be processed, and that the resulting music is an interpretation has understood the essential science. The sound is not the brain speaking in a hidden language. It is a carefully constructed way of making neural data perceptible — one that works only when its technical limits remain part of the performance.