The Rapid Emergence of Musical Pitch Structure

in Human Cortex
Analysis by Paula Pérez Romero
KU LEUVEN
Introduction
We are constantly inferring meaning from the music and noise that surround us. Tonal
music gives us the framework for examining this interaction. There are different tonal systems, in
which the pitch is arranged hierarchically in different ways that make some pitchs to be perceived
more stable than other. The Western tonal music has being quantified as the standard tonal
hierarchy (STH). Neural substrates supporting STH are still highly misundestood.
This article tries to answer two important inquiries, firstly, “do cortical population encode
musical pitch in a manner that precipitates the organization of the STH?” (Sankaran et al, 2020)
and secondly, “what are the representational dynamics underlying the emergence of such a
perceptual representation from lower-level afferent information?” (Sankaran et al, 2020).
For doing so, cortical activity was recorded via MEG while music experts listened to different
pitch music. Several analysis that I will further explain in depth were applied.

Materials and Methods

Apparatus
Data were collected with a MEG system. I think that the authors choose this method and
not other due to several reasons. Firstly because this method provides an excellent temporal
resolution, which will allow the researchers to capture precise time of neural activity, something
crucial when studying auditory processing in the context of music as the pitch perception occurs
very fast. It also gives a non-invasive, direct measurement of neural activity while covering the
whole-brain, useful as it allows us to see the different regions involved in integrating the
information. And finally, it is very sensitive to neural oscillations, which allows investigators to
investigate the frequency-specific dynamics of neural activity associated with the perception of
musical pitch.
The MEG data were filtered to focus on the frequency range of interests using first-order
RC filter and subsequently digitised at a sampling rate of 1000 Hz for further analysis. This
filtering is useful for removing unwanted noise and artefacts.

Stimuli
Trials are composed of tonal context followed by 1 of 12 different tones. Before the test,
tones were passed through a time-varying-loudness model and normalise for differences in
perceived loudness using STLmax, which helps account for individual differences in loudness
perception, standaridize loudness levels across stimuli as well as enhance comparability of neural
responses and a better mimic of the natural auditory perception of musical tones. All of these
features contribute to a more accurate investigation of the cortical representation of musical pitch
by focusing on the specific perceptual attribute of interest while minimising confounding factors.

1
Also to prevent sensory processing of the context from contaminating evoked responses to prove
tones, tones were separated by a silent period.

Experimental design
By recording MEG activity from subjects as they listened to different tones and performed
a behavioral task the researchers were able to examine the neural dynamics and decoding accuracy
associated with different aspects of pitch perception. This experimental desing has influenced the
result in various ways.
To begin, it is important to highlight that participants were engaged in a task where they
had to judge wheter the probe tome was “in-key” or “out-of-key”. By being asked to complete
this task, it was ensured that participants were attending to the stimuli and fully focused on pitch
perception. Also, it also helped to provide an insight into their perceptual judgments as the
accuracy of their responses was measured and analysed.
In line with this, I would like to mention that the motor activity was also control by
switching the “in-key” and “out-key” responses to left and right bottoms every two blocks. This
control helped to make sure that neural differences are not solely due to motor responses but they
are indeed related to the perception of the pitch.
The way in which tones are presented is also important. In this experiment, they were
presented in a random order without repeats across trials, which helped to control for potential
order effect and ensures that the results are not biased by the specific order of tone presentations.
Another key factor of this experimental design is the inclusion of incorrect responses,
which allows a comprehensive examination of the neural activity associated with different
perceptual judgements regardless of their accuracy, so it provides a more complete picture of the
neural representation of pitch.
Lastly, the used of magnetoencephalography to record the neural activity of participants
has several highlights that were outlined in a previous section (see Apparatus)

Analysis
The analysis of the data consisted in 3 parts: MEG preprocessing, performed via MATLAB;
MVPA; and RSA.
1. MEG preprocessing: firstly the neural epochs were extracted from 100 ms before to 1000
ms after the onset of probe tone. Then the data were downsampled to 100 Hz to reduce
computational load while preserving relevant information. A low-pass Chebyshev Type
1 filter was applied to remove high-frequency noise and artifacts. Afterwards, PCA was
used to reduced the dimensionality of the data. It identified the principal component that
explained 99% of the variance in the dataset. The number of features was reduced from
160 channels to an average of 28 principle components.
This preprocessing process is important for assesing the quality of the data and the impact
of data transformation on subsequent analyses.
2. MVPA: binary classifiers attempted to decode the identity of probe tones from recorded
brain activity. To enhance the signal-to-noise ratio there were used averaged responses of
two exemplars within each class. Then, the Naive Bayes linear discriminate analysis
(LDA) was used for classification. Its generalización was evaluated using k-fold cross-
validation and the reported accuracy was an average across all folds. The use of a sliding
window approach allowed the assessment of decoding accuracy based on spatial
activation patterns ans on the local temporal structure of neural responses. And lastly, the
classifier performance at each time point was evaluated using balanced accuracy.

2
 Having a good understanding of MVPA methodology is crucial for interpreting the
reported decoding accuracies, as researchers can evaluate the reliability ad
generalisability of the classification results and assess the temporal dynamics of neural
information processing.
3. RSA: the MVPA was applied to each pairwise combination of probe tones resulting in a
neural representational dissimilarity matrices (RDMs) that were later compared with
different model RDMs based on perceptual and sensory models of pitch. This approach
allowed the researchers to compare empirical neural representation with model-based
representation, which can help in gaining insights in the neural coding of perceptual and
sensory dimensions of pith .
Each of the three step has signficance in ensuring data quality, understanding neural
information processing dynamics and investigating the neural coding of perceptual and sensory
dimensions.

Control experiment
The control experiment introduces important manipulations and analyses that allowed
researcher to evaluate the impact of certain factors on neural decoding and representation.
In this control experiment, participants were instructed to perform a timbre-identification
task instead of the pitch analysis task used in the main experiment. This manipulation aims to
assess the impact of the behavioral task on neural decoding. By comparing the results from the
control experiment with those from the main experiment, researchers could determine the extent
to which neural decoding is influenced by the specific task demands related to pitch perception.
This control condition provided valuable insights into the neural processing of pitch independent
of explicit pitch analysis.
Another manipulation that was introduced in this condition is the alteration of the
statistical distribution of pith in the context, which allowed scientists to compare the neural
responses to probe tones in these different contextual condition and helped then assess how the
statistical distribution of pitch in the context impacts the neural representation of pitch in
subsequent stimuli.

The analysis of the control experiment followed a similar procedure to the main
experiment, with some modifications to account for the specific manipulations. Preprocessing
steps were performed similarly, but neural data corresponding to trials with flute tones were
discarded from the analysis. Additionally, because of the lower number of subjects and trials in
the control experiment, adjustments were made to classification parameters and downsampling
was performed to maximize the signal-to-noise ratio. The analysis also included the examination
of candidate representational dissimilarity matrices (RDMs) based on pitch class probability, f 0
probability, and statistical overlap (SO) to investigate the relationship between the statistical
distribution of pitch in the context and the structure of dissimilarities between probe tones. This
data analysis is important because it helps to isolate and evaluate specific factors, validates the
main experiment’s results and provides insights into the neural mechanisims underlying pitch
perception, which together helps to draw more accurate conclusions and strengthens the overall
scientific interpretation of the study.

To assess whether the structure of neural RDMs reflected gradual adaptation of MEG
responses to stimulus regularities over the course of the experiment, an analysis was performed
by applying MVPA repeatedly on four-block windows of adjacent data. This analysis helps to

3
understand how neural representations evolve over time and adapt to the stimulus regularities
presented throughout the experiment.
In general, this control condition helped to validate the findings from the main experiment
and strengthened the reliability and the specificity of the results.

Results
The researchers found that neural activity patterns in the brain could accurately
distinguish between different musical pitches. By using MVPA, they were able to decode the
specific pitch classes based on the patterns of neural activity, which suggests that the brain
represents and discriminates musical pitch in a precise and distinguishable way.
It was also found that the brain gradually adapted its neural responses to the recurring
patterns in the stimuli over the course of the experiment, suggesting that the brain dynamically
adjusts its representations and responses to match the regularities present in the musical stimuli,
indicating the involvement of learning processes and neural plasticity in pitch perception.
Overall, the results demonstrate that the human brain represents and processes musical
pitch in a precise and distinguishable manner. The findings provide insights into the neural basis
of pitch perception and contribute to our understanding of auditory perception and music
cognition. The rigorous methodology and control conditions employed in the study enhance the
confidence in the reported results and their implications for our understanding of how the brain
perceives and processes music.

Discussion
The discussion emphasises the significance of the results and raises important avenues
for future research.
The study’s rigorous methodology, including the use of temporal decoding techniques
and control experiments, strengthens the validity and reliability of the findings. By bridging the
gap between music perception and neuroscience this research contributes to our understanding of
how the brain processes and represents music, with potential implication for both, music cognition
and reader perceptual processes.

References:

Sankaran, N., Carlson, T. A., & Thompson, W. R. (2020). The Rapid Emergence of
Musical Pitch Structure in Human Cortex. The Journal of Neuroscience, 40(10),
2108–2118. https://doi.org/10.1523/jneurosci.1399-19.2020

Peretz, I., & Zatorre, R. J. (2005). Brain Organization for Music Processing. Annual
Review of Psychology, 56(1), 89–
114. https://doi.org/10.1146/annurev.psych.56.091103.070225

De Beeck, H. O., & Nakatani, C. (2019). Introduction to Human