An EEG trace follows the rhythm of a repeating tone. A graph shows a sharp peak at the tone rate, and the result is called neural entrainment. What has actually been measured: an internally sustained oscillation aligning to the sound, or a succession of responses evoked by each tone? Both can generate a periodic scalp signal. Measuring neural entrainment requires an explicit account of the stimulus, the neural response and the inference that connects them.
Researchers use “entrainment” in more than one sense. At a descriptive level it can mean that neural activity is temporally aligned with a repeating external event. A stronger mechanistic claim is that an endogenous neural oscillator changes its phase or frequency in response to that event. The latter cannot be read directly from a single spectral peak. This Methods & Measures guide explains why frequency tagging, phase alignment, attention effects and continuation after stimulation answer different questions. It is distinct from measuring beat or meter as a conscious musical percept.
Define the proposed coupling
A rhythmic stimulus has physical periodicities: onsets, amplitude modulation, accents and often slower phrase structure. The brain’s recorded response may contain corresponding periodicities. A useful first question is therefore whether the response has energy or phase consistency at a frequency present in the stimulus. A second asks whether the brain represents a periodicity inferred by the listener even when it is weak or absent acoustically. A third asks whether an intrinsic rhythm was altered by the stimulus. These claims become progressively harder to establish.
The Nature Reviews Psychology synthesis by Snyder, Gordon and Hannon describes behavioral and neural work on musical rhythm, beat and metre. It supplies context but does not validate every use of an EEG entrainment score. A measurement report should name the frequency being tested, why that frequency matters and what comparison supports the interpretation. If a beat rate is 2 hertz and a grouping rate is 1 hertz, those components should not be treated as one undifferentiated “rhythm response.”
The word “neural” also needs scale. Scalp EEG measures a mixture of electrical signals generated by populations of neurons, filtered by tissue and recording hardware. It does not observe a single cell locking to each beat. Magnetoencephalography and intracranial recordings offer different spatial and temporal tradeoffs. A claim about “the brain synchronizing” is acceptable shorthand only if the measured signal and limitations are made explicit.
State whether the study is about auditory rhythm, speech envelope tracking, visual flicker or another domain. Similar analyses may be used across them, but the stimulus-to-brain pathway and the meaning of a response differ. A speech envelope can contain irregular syllabic fluctuations rather than one stable pulse. A tone train has clean periodic events but may be less like ordinary listening. Neither paradigm automatically generalizes to all kinds of temporal coordination.
Compare input and output directly
Frequency tagging deliberately presents a periodic stimulus and looks for a response at the same frequency or its harmonics. Nozaradan and colleagues used this approach in EEG studies of beat and meter. The method’s strength is a specified input-output comparison. Its vulnerability is the temptation to call every output peak an internally generated metrical representation, even when the sound itself carries that periodicity.
Compute the spectrum or temporal structure of the actual delivered stimulus, not just the composition file. Speakers, headphones, room acoustics and compression can add or alter modulation. A complex rhythm may have strong energy at multiple frequencies, including subdivisions and harmonics. If the purported neural meter peak matches an acoustic feature that was not inspected, the stronger endogenous claim is unsupported. Show input and output plots on compatible frequency axes.
In their 2011 work, Nozaradan and colleagues asked listeners to imagine binary or ternary meter over a beat and examined corresponding EEG frequencies. The attention or imagery manipulation helps separate listener organization from physical repetition. Their later work with more complex rhythms compared selective responses to beat and meter components. These are useful experimental contrasts; they do not turn one participant’s spectral amplitude into a universal capacity score.
Auditory pathway responses can also change across levels of processing. Nozaradan, Keller and Lenc examined cortical and lower-level responses to rhythms and argued that selective enhancement of meter-related frequencies is not fully explained by lower-level acoustic properties. The inference depends on their recording and contrast, not on every frequency-tagging study. A new experiment needs its own stimulus controls and signal-quality evidence.
When the stimulus is not strictly periodic, time-domain alignment may be more informative than one average spectrum. Researchers may correlate neural activity with an envelope or estimate a temporal response function. Obleser and Kayser distinguish neural entrainment from the more pragmatic signature often called speech tracking. Tracking shows a reproducible relationship between stimulus and recorded activity; the mechanism can still involve evoked responses, predictive processes or oscillatory dynamics.
Do not confuse periodic response with mechanism
A series of identical tones can evoke a response to each onset. Summing those responses produces a periodic waveform and a spectral peak. An oscillator that adjusts its phase to the tones can produce a similar observed pattern. Zoefel, ten Oever and Sack describe the difficulty of separating endogenous oscillations from repeated evoked responses and review experimental signs that may sharpen the inference. A scalp average alone cannot settle the debate.
One strategy is to vary the regularity of the input while matching the number, intensity and timing distribution of events as closely as possible. Another introduces an omission where a beat was expected. A neural response around the omitted event may suggest an internal prediction, although anticipation, attention and the surrounding acoustic context must still be considered. A continuation effect after the sequence ends can be informative, but the analysis must distinguish neural persistence from filter ringing and task-related anticipation.
Phase is often central to an entrainment claim. If a neural oscillation is aligned to a stimulus, phase relationships should be reported with their variability. A significant phase-locking value can arise from repeated event-locked responses. Compare against physically and statistically appropriate controls, and specify whether phase is measured relative to event onsets, the acoustic envelope or a participant’s chosen beat. Different references answer different questions.
Power and phase measures should not be treated as interchangeable. A condition can increase evoked response amplitude without improving temporal alignment, or alter phase consistency without a large average power change. Baseline subtraction, filtering and time-frequency windows can shift the apparent effect. Prespecify the outcome or mark exploratory analyses clearly. A selected peak among many frequencies and electrodes has a higher false-positive risk than a targeted, independently motivated frequency.
Haegens and Zion Golumbic critically review claims that rhythm facilitates sensory processing. Their caution is relevant because a behavioral advantage after rhythmic stimulation does not by itself prove a specific entrainment mechanism. Attention, expectation and sensory adaptation can improve performance. Use the word entrainment descriptively when the evidence is temporal correspondence, and reserve stronger mechanistic language for designs that test alternatives.
The pattern of trial averaging also changes the claim. An evoked response that occurs at a consistent delay after every sound can look especially strong after averaging across presentations, while a rhythm that varies in phase from trial to trial may disappear in that average despite being present in single trials. Examine single-trial and averaged estimates where feasible, and describe how phase and amplitude were combined. Zoefel and colleagues emphasize that tests for endogenous involvement depend on predictions that repeated evoked-response accounts would not make. Thus a study should articulate the competing models before selecting a statistic, then report whether the observed result actually discriminates between them.
Keep the participant’s behavior in view
EEG is not a direct report of what the listener heard. Ask participants about beat location, grouping or timing when those experiences are central. A brain response at a meter frequency may occur during directed attention without a stable conscious meter judgment. Conversely, a person may report a clear beat with a weak scalp peak because of noise, anatomy or the selected analysis window. Behavioral and neural outcomes should be reported side by side rather than used as substitutes.
Movement can contaminate the signal and change perception. Finger tapping, head sway and facial muscle activity can create periodic electrical or mechanical artifacts. Yet forbidding all movement may change the listening experience. If participants must remain still, say so and interpret the result under that constraint. If movement is allowed, record it and test whether a frequency peak tracks muscle or motion rather than neural activity of interest.
Attention is another manipulable factor. A person instructed to attend every second beat may show a different response from one attending every third beat over identical acoustic input. That is valuable for testing top-down organization. It also means that a difference between conditions can reflect task instruction rather than a stable trait. Report comprehension checks and the timing of instruction relative to stimulus presentation.
Musical training, cultural familiarity, age, hearing and fatigue can alter both neural and behavioral responses. A particular rhythm may be easy for one sample and opaque for another. If an individual score is intended, test whether it remains stable across days, stimuli and instructions. The reproducibility of a group-level effect does not guarantee that every participant can be ranked reliably on a neural entrainment trait.
Plan acquisition and preprocessing for the claim
Report the EEG or MEG instrument, sampling rate, reference scheme, electrode or sensor selection, filtering and artifact handling. Temporal precision matters when the effect is defined by a phase relation. Filters can smear activity before and after a stimulus event. If a post-stimulus continuation is the key evidence, test the filter’s impulse response and include a control demonstrating that the apparent continuation is not introduced by preprocessing.
Length of stimulation determines frequency resolution. A short sequence may be adequate to show event-locked responses but less suitable for separating nearby beat and meter frequencies. Many repeated trials improve the average while potentially increasing habituation or strategy changes. State how many usable trials remained for each participant and whether rejected trials differed between conditions.
Choose a noise baseline carefully. A narrow spectral peak should be compared with neighboring bins or an appropriate control, but neighboring bins can contain leakage from the same finite-duration signal. Window functions, detrending and spectral smoothing affect peak size. A raw amplitude plot is insufficient if the analysis pipeline is undisclosed. Provide enough information for another team to reproduce the input-output comparison.
For group inference, account for participants rather than treating every sensor and trial as independent. Multiple electrodes can be informative, but selecting the strongest after inspecting the effect inflates estimates. A preregistered region or cross-validated selection is preferable when making a confirmatory claim. Replication across independent samples and rhythm sets matters more than a highly optimized effect in one dataset.
Finally, compare the neural signature with a control that tests the intended psychological inference. If the claim is meter perception, vary imagined or judged meter over acoustically identical material. If the claim is attention, direct attention while keeping the sound constant. If the claim is oscillator persistence, examine activity after omitted or stopped events with artifact controls. A peak at the stimulus rate alone answers the narrower question of periodic response.
What the evidence does not show
The cited research does not establish that every periodic EEG response is a self-sustaining neural oscillator. Repeated evoked responses and stimulus tracking can create similar patterns. Nor does a larger spectral peak show that a person understands music more deeply, is more attentive in general or has a healthier brain. The laboratory contrast and signal-processing pipeline determine what can be concluded.
The beat and meter EEG studies demonstrate informative response differences under specific stimuli and tasks. They do not supply a clinical cutoff, a universal individual score or a way to read all metrical experience from one scalp electrode. The speech-tracking literature concerns related but not identical signals and cannot be imported as proof for a musical beat paradigm without matching design.
The mechanistic debate remains active. A missing-beat response, phase alignment or post-stimulus oscillation can narrow alternatives when carefully controlled, but none is automatically decisive. The safest report states the observed neural-stimulus relation and separately grades the evidence for endogenous entrainment.
Make the inference visible
A reader should be able to reconstruct the sound’s timing and spectrum, the participant’s task, the acquisition and artifact rules, the neural frequency or phase metric, the control comparison and the behavioral result. Show whether the response followed a physical feature, an instructed grouping or a self-reported percept. If the article’s opening tone train produced a peak, that peak is a beginning: it establishes a temporal relationship to explain, not the end of the explanation.
Neural entrainment is most useful as a precise question about coupling across time. Measurement earns its value when it preserves the distinction between a signal that follows a rhythm and a mechanism that generates expectations about it. That distinction lets a striking graph become a reliable finding.
Research references
- Snyder, J. S., Gordon, R. L., & Hannon, E. E. (2024). Theoretical and empirical advances in understanding musical rhythm, beat and metre. Nature Reviews Psychology, 3, 449–462. View source
- Nozaradan, S., Peretz, I., Missal, M., & Mouraux, A. (2011). Tagging the neuronal entrainment to beat and meter. Journal of Neuroscience, 31, 10234–10240. View source
- Nozaradan, S., Peretz, I., & Mouraux, A. (2012). Selective neuronal entrainment to the beat and meter embedded in a musical rhythm. Journal of Neuroscience, 32, 17572–17581. View source
- Nozaradan, S., Schönwiesner, M., Keller, P. E., Lenc, T., & Lehmann, A. (2018). Neural bases of rhythmic entrainment in humans: critical transformation between cortical and lower-level representations of auditory rhythm. European Journal of Neuroscience, 47(4), 321–332. View source
- Zoefel, B., ten Oever, S., & Sack, A. T. (2018). The involvement of endogenous neural oscillations in the processing of rhythmic input: more than a regular repetition of evoked neural responses. Frontiers in Neuroscience, 12, 95. View source
- Obleser, J., & Kayser, C. (2019). Neural entrainment and attentional selection in the listening brain. Trends in Cognitive Sciences. View source
- Haegens, S., & Zion Golumbic, E. (2018). Rhythmic facilitation of sensory processing: a critical review. Neuroscience & Biobehavioral Reviews, 86, 150–165. View source