Research on the Effect of Background Music on Working Memory Based on Granger Causal Network

Working memory is the basis of many advanced cognitive functions which has become an important research content in many fields. Working memory refers to the memory system that temporarily processes and stores information when carrying out complex cognitive activities, such as learning, language, reasoning. Nowadays, lots of people like reading and studying with some background music. Music is the outcome of human civilization and the media for people to transmit information and express their emotions. Existing studies show that music has effects on people's cognitive function.

In 1993, Felix used rock music, romantic music and no-music to conduct instant recall experiments. He found the result with romantic music was the best and the result without music was worst ^[1]. In 1995, Rauscher et al. found that after listening Mozart's music, the students’ intelligence test results got higher than others, which was named after Mozart Effect ^[2]. In 2000, Anderson asked his subjects to memorize numbers under pure music background and songs background respectively and found that the performances under pure music background were higher ^[3]. In 2004, Goldstein’s group asked two groups of children to learn words and numbers in classical music background and quiet background respectively^[4]. The final result showed that the auditory memory of children in classical music group was significantly better than that in quiet background group.

These studies show that music can affect working memory, and different types of music shows different effects. On the one hand, Researchers believe that soothing music can relieve people's tension, relax their mood and eliminate anxiety, while help improve the memory and learning efficiency. Some music may induce the potential huge power in people's mind, combine scattered and isolated information in the mind, and promote the creative thinking. On the other hand, some researchers think that when memory, learning tasks and background music exist at the same time, background music is an irrelevant stimulus, occupies the cognitive resources in brains and leads to bad learning performances ^[5].

EEG is a kind of complex and unstable random signal measured from the scalp of the brain, which is the external manifestation of brain nerve activity, and contains rich physiological and disease information. The nonlinear dynamics of EEG signals shows that there is a nonlinear and coupling relationship between different brain regions, which introduces a research direction based on brain network to explore working memory.

At present, many researchers have explored the brain network, mostly using magnetic resonance imaging(MRI), EEG, magnet encephalograph(MEG) and so on. Through the correlation or causal analysis of collected EEG signals, and the analysis of network characteristic parameters, the topological structure of the brain network is studied, so people can explore the working mechanism of the brain. Micheloyannis et al. collected EEG signals in resting state and working memory state of schizophrenics and normal group respectively, and conducted brain network analysis. The result showed that compared with the normal group, the "small world attribute" of brain function network of schizophrenics was abnormal ^[6]. Stam et al. conducted brain network research on EEG signal data of Alzheimer's patients, and found that the small world characteristics of brain function network in Alzheimer's patients were lost ^[7]. Haan et al. compared the EEG signal data of Alzheimer's patients with normal ones and found that the brain function networks of patients group were abnormal ^[8].

Norbert Wiener first introduced the concept of causality in 1956^[9]. In 1969, Clive Granger put forward a method to measure causality between time series by using linear regression model of random processe^[10]. In 2013, palaniyappan et al. used Granger causality model to study fMRI data of schizophrenics in resting state and found that the positive feedback and interaction between the insula and dorsolateral prefrontal cortex in schizophrenic patients changed to a great extent, and the connection between bilateral visual cortex and insula was abnormal ^[11]. In 2014, Maria et al. used the Granger causality analysis method to study the auditory hallucinations of schizophrenics and found that neural connections of patients with auditory hallucinations were abnormal under audio stimuli^[12].

In this paper, we first introduce a designed working memory experiment and necessary methods. Then the experiment results and the process of building the causal networks are represented. The analysis of experiment data and the characteristics of the causal networks is made. Finally, we draw our conclusions.

2

Materials

2.1

Experimental Subjects

Twenty students without musical training were randomly chosen. They are all right-handed, in good health and with normal or corrected vision and normal hearing function.

2.2

Experimental Materials

In this research, experimental materials consist of English words and background music. The selected English words are low-frequency words in IELTS dictionary and all professional words, which reduces the impact of individual differences.

Classical and rock music were selected. Classical music is usually slow, gorgeous and smooth; on the contrary, rock music has a strong, distinct and passionate rhythm. For classical music, Mozart's Moonlight Sonata is chosen. For rock music, Last ride of the day created by a Finnish band, Nightwish, is chosen. Both of them are without lyrics.

2.3

Experimental Equipment

The acquisition equipment in this experiment is the EEG experiment recorder produced by NeuroScan company in the United States. It is equipped with a 64-electrode cap and a SynAmps amplifier. The acquisition software is Scan system, and the sampling frequency is 1kHz.

2.4

Experimental Design

Sternberg memory paradigm is widely used in the study of working memory, which uses memory items to detect the level of human memory. In the traditional Sternberg working memory experiment, the memory item to be memorized is required to appear one by one. At the same time, subjects need to memorize all the items initially proposed and compare them with a probe item. In this research, an improved Sternberg working memory task experiment is adopted. All the memory items appear in front subjects of at once and disappear after a certain time, so that more ideal EEG data can be collected.

The experiment flow needs to be introduced to subjects first. The order of three kinds of background music is no music, classical music and rock music. Subjects are required to memorize 8 English words in 40 seconds under the condition of the background music. When the stimulation finishes, subjects enter the memory retention stage with a duration of 10 seconds. Then the probe item appears in the center of the screen one by one. Subjects need to judge whether the probe item has appeared before and push the mouse button to react. Right button for ‘Yes’ and left button for ‘No’. This stage lasts 120 seconds. The above is a complete task that is shown in the figure 2.1. Under each condition, subjects need to do the task three times. There are 5 seconds between tasks under the same condition and 5 minutes for subjects to rest between conditions to reduce the influence of former stimulation. During the whole experiment, the EEG data of subjects are collected.

3

Methods

3.1

Preprocessing

The preprocessing of the EEG data contains such steps:

Electrooculogram (EOG) correction. Collect the EOG signals synchronously and correct other signals affected by the EOG with the electrooculogram as the positive and negative reference.

Ignore bad channels. After this operation, there are 49 valid channels left. Remove the EEG data of bad channels.

Digital filtering. Filter the EEG data with a band-pass filter whose passband is 1Hz to 40Hz.

Set reference electrode. In this research, the infinity reference based on reference electrode standardization technique (REST) is used. This approach solves the problem that there is no neutral point on the surface of the head by transferring the reference electrode to infinity^[13-15]. The direct EEG problem can be solved by the following formula: 1 $V = G S$ \[V=GS\]

In this formula, G is the transfer matrix with the reference at infinity and depends on the shape of head model, the properties of source and the type of electrodes; S is the source; V is the EEG data with the reference at infinity.

Segmentation of the EEG data. Make segments according to the time points of the stimulation and 1.2s before and 2s after the responses of subjects.

Independent Component Analysis (ICA). ICA can separate signals that are statistically independent but linearly mixed.

Remove artifacts from head movement, swallowing, linear drift, etc.

3.2

Analysis of behavior data

3.2.1

Grubbs rule

The behavior data in the experiment can be expressed by the response time and accuracy. During the tasks, the behavior data can be synchronously collected. The Grubbs rule was used to eliminate the abnormal parts of the behavioral data.

The basic ideal of Grubbs rule is that in a set of repeated observations x_i, x_d has the largest absolute value of the residual. The supposed confidence probability is P = 0.99 or P = 0.95, which means the significant level is a = 0.01 or a = 0.05. $If \frac{| x_{d} - \bar{x} |}{s} \geq G (a, n) x_{i}, x_{d}$ \[~\text{If}~\frac{\left| {{x}_{d}}-\bar{x} \right|}{s}\ge G(a,n){{x}_{i}},{{x}_{d}}\] can be judged as abnormal, namely gross error.

3.3

Granger causal network

3.3.1

Granger causality

Granger causality is based on the multiple autoregressive model and its basic idea is that X and Y are two variables and the past values of X and Y are used to predict the future value of X. If the effect is better than what we get with the past value of X alone, it will indicate that the past value of Y contains useful information to predict the future value of X. This information is not included in the past value of X, so that we can conclude that there is a causality between X and Y^[16-17].

The principle of Granger causality analysis based on autoregressive equation is as follows^[18].

Suppose X(t) and Y(t) are two stable time vectors and their linear autoregressive equations are expressed here: 2 ${\begin{array}{l} X (t) = \sum_{i = 1}^{p} a_{1 i} X (t - i) + ε_{1 t} \\ Y (t) = \sum_{i = 1}^{p} b_{1 i} Y (t - i) + ε_{2 t} \end{array}$ \[\left\{ \begin{array}{*{35}{l}} X(t)=\underset{i=1}{\overset{p}{\mathop \sum }}\,{{a}_{1i}}X(t-i)+{{\varepsilon }_{1t}} \\ Y(t)=\underset{i=1}{\overset{p}{\mathop \sum }}\,{{b}_{1i}}Y(t-i)+{{\varepsilon }_{2t}} \\ \end{array} \right.\]

where ε_1t and ε_2t are the errors of estimation, a_1i and b_1i are model fitting parameters and p is the order.

According to the equations, we can get the linear regression equations under two conditions that X(t) is affected by the past information of Y(t), and that Y(t) is affected by the past information of X(t) separately.

3.3.2

Directional transfer function

DTF is a quantity that describes the direction and intensity of information transmission between signals of each channel, through which the strength of causality between channels can be judged^[19-20]_.

Suppose the multi-channel EEG signals in time domain are: 3 $x (t) = {[x_{1} (t), x_{2} (t), \dots, x_{N} (t)]}^{T}$ \[x(t)={{[{{x}_{1}}(t),{{x}_{2}}(t),\ldots ,{{x}_{N}}(t)]}^{T}}\]

where N is the number of channels and t is the time. Based on Multivariate autoregressive(MVAR) model, x(t) can be fitted to the data as that: 4 $\sum_{k = 0}^{P} λ (k) X (t - k) = E (t)$ \[\underset{k=0}{\overset{P}{\mathop \sum }}\,\lambda (k)X(t-k)=E(t)\]

where E(t) is the multi-parameter zero-mean white noise vector, the elements of λ(k) are the parametric matrixes of MVAR model, λ(0)=I and p is the order. In order to explore its spectral characteristics, the above formula can be transformed into: 5 $λ (f) X (f) = E (f)$ \[\lambda (f)X(f)=E(f)\]

where $λ (f) = \sum_{k = 0}^{p} λ (k) e^{- j 2 π f ΔΔ t}$ $\lambda (f)=\underset{k=0}{\overset{p}{\mathop \sum }}\,\lambda (k){{e}^{-j2\pi f\Delta \Delta t}}$ and f is the frequency. The above formula can be also transformed into: 6 $X (f) = λ^{- 1} (f) E (f) = H (f) E (f)$ \[X(f)={{\lambda }^{-1}}(f)E(f)=H(f)E(f)\]

Where H(f) is the reciprocal of conversion coefficient matrix in frequency domain and λ(f) is the transformation matrix of the system. So the DTF matrix which describes the strength of the causal connection between channel i and channel j can be constructed as: 7 $γ_{i j}^{2} (f) = \frac{{| H_{i j} (f) |}^{2}}{\sum_{m = 1}^{N} {| H_{i m} (f) |}^{2}}$ \[\gamma _{ij}^{2}(f)=\frac{|{{H}_{ij}}(f){{|}^{2}}}{\underset{m=1}{\overset{N}{\mathop \sum }}\,|{{H}_{im}}(f){{|}^{2}}}\]

where $γ_{i j}^{2} (f)$ $\gamma _{ij}^{2}(f)$ is the normalized result and presents the proportion of the influence of channel j on channel i in the influence of all channels on channel i. The larger the value is, the stronger the causality connection between channel j and channel i is. After the DTF matrix was constructed, the average causal connection strength can be obtained by computing the average of the matrix, which was used to describe the strength of causal connections between signals.

The expression of the average causal connection strength of the whole brain area is as follows: 8 $D T F_{m e a n} = \frac{1}{N (N - 1)} \sum_{i \in K} \sum_{j \neq i \in K} D T F_{i j}$ \[DT{{F}_{mean~}}=\frac{1}{N(N-1)}\underset{i\in K}{\mathop \sum }\,\underset{j\ne i\in K}{\mathop \sum }\,DT{{F}_{ij}}\]

Where DTF_ij is the DTF value from channel j to channel I, and presents the edge of the causal network; N is the number of electrodes used, and it presents the number of the nodes of this network; K is the set of all the nodes. DTF_mean value describes the strength of the network connection. The bigger the value is, the stronger the network connection is.

3.3.3

Construction of causal network based on DTF matrix

The construction process consists of three steps: the selection of nodes, the definition of edges and the binarization. Different selections have huge impact on the structure of the network.

1

selection of nodes

The scalp electrodes are usually chosen as the nodes to construct the brain network. The current international standard EEG recording system is 10-20 system. In this system, the locations of the 64 electrodes are shown in fig 3.

2

Definition of edges

The common methods to establish edges of networks are based on cross correlation, mutual information, synchronous likelihood method or DTF, etc. In this paper, the DTF matrixes of alpha band and beta band under three conditions were computed respectively and the value DTF_ij are used to define the edge E_ij.

3

Binarization

The DTF matrixes need to be binarized with an appropriate threshold. There is a principle that the average degree of the nodes should be greater than the natural logarithm of the number of the nodes $\bar{k} \geq 2 \ln N;$ $\bar{k}\ge 2\ln N;$ meanwhile, the effect connection sparsity S is less than 50%. Therefore, the threshold T can be kept in the range 0.01 ≤ T ≤ 0.9 and the binarization is performed by the increment of 0.01.

After three steps above, the causal networks are finished.

3.3.4

Analysis of topological parameters

1

node degree

The value of degree of some node represents the number of its neighbor nodes in the network. The degree K_i of node i can be computed by the following formula: 9 $K_{i} = \sum_{i, j \in V} a_{i j}$ \[{{K}_{i}}=\underset{i,j\in V}{\mathop \sum }\,{{a}_{ij}}\]

where V is the set of the nodes in the network, a_ij is the element of the binarized DTF matrix. a_ij = 1 represents one connection between node i and node j.

2

global efficiency

The global efficiency is used to express the information transmission speed in the network. Global efficiency is the average of the reciprocal of the shortest paths of all nodes and can be used to measure how fast the brain network transmits and processes information. Global efficiency E_global is expressed as follows: 10 $E_{g l o b a l} = \frac{1}{N (N - 1)} \sum_{i, j \in V, i \neq j} \frac{1}{l_{i j}}$ \[{{E}_{global~}}=\frac{1}{N(N-1)}\underset{i,j\in V,i\ne j}{\mathop \sum }\,\frac{1}{{{l}_{ij}}}\]

where N is the number of nodes, l_ij is the shortest path from node i to node j. The more larger E_global is, the stronger the global information transmission and processing ability of the network is, and the lower the cost of information transmission is.

3

connection density

Connection density D refers to the density of connections between nodes in a network. Its expression is as follows: 11 $D = \frac{1}{N (N - 1)} \sum_{i \neq j, i, j \in V} γ_{i j}$ \[D=\frac{1}{N(N-1)}\underset{i\ne j,i,j\in V}{\mathop \sum }\,{{\gamma }_{ij}}\]

Where N is the number of nodes, γ_ij is the connection value of node i and node j. The range of connection density is 0 ≤ D ≤ 1.

4

information flow gain

Information flow gain is proposed on the basis of information flow. Information flow is divided into inflow information and outflow information. Inflow information refers to the sum of information received by one channel from other channels, while the other refers to the sum of information output from one channel to other channels. Their expressions are as follows: 12 $f l o w_{I N} = \sum_{j = 1}^{N} D T F_{m j}$ \[~flow{{~}_{IN}}=\underset{j=1}{\overset{N}{\mathop \sum }}\,DT{{F}_{mj}}\] 13 $f l o w_{O U T} = \sum_{i = 1}^{N} D T F_{i m}$ \[~flow{{~}_{OUT}}=\underset{i=1}{\overset{N}{\mathop \sum }}\,DT{{F}_{im}}\]

The information flow gain is expressed as follows: 14 $ρ_{m} = f l o w_{O U T} / f l o w_{I N}$ \[{{\rho }_{m}}=flow{{~}_{OUT}}/flow{{~}_{IN}}\]

Where ρ_m indicates the contribution of channel m in the process of information interaction.

4

Results

4.1

Behavioral data

The accuracies and the response time can be computed with the behavior data. The accuracy refers to the percentage of correct response in the total number of times when the subjects give the response in the working memory tasks. The response time is the item from the appearance of detection items to the response of the subjects during the tasks. The average accuracies under three conditions are shown in figure 1.

As is shown in the figure, the mean accuracies with music are lower than the mean accuracy without music. The mean accuracy under with rock music is the lowest. One-way analysis of variance (ANOVA) is used on the data, which shows that the accuracies without music are higher than the mean accuracies with classical music and rock music, and the difference is significant (P=0.049, P<0.05). The accuracies under the condition of classical music are higher than the mean accuracies under the condition with rock music, and the difference is significant (p=0.004, p<0.05).

We screened all the recorded response time data and computed their averages under three conditions. The result is shown in fig 2.

It shows that subjects spend longest time to make their responses with rock music and shortest time without music. ANOVA is used on the data, which shows that the mean response time without music is shorter that with classical music and rock music, and the difference is significant (P=0.001, P<0.05). The mean response time with classical music is shorter than that with rock music, and the difference is significant (P=0.003, P<0.05).

4.2

EEG data

According to the principle of symmetrical location of left and right brain electrodes, the data from 32 channels are selected. They are from frontal lobe, parietal lobe, temporal lobe and occipital lobe.

In this paper, alpha-band and beta-band data were selected to construct the brain network. The EEG data in alpha band and beta band of 20 subjects are pre-processed and the average DTF matrixes under three conditions are computed. The mean DTF values are shown in Tab 1. The result shows that compared with those without music, the mean DTF values of the subjects with classical music and rock music decrease, which indicates that the causal connection strength between the nodes is weakened, and the information transmission ability between the brain regions is weakened. Meanwhile, T-test is used to compare the difference between the average DTF values under different conditions of the same frequency band. In alpha band, there is, a significant difference (P=0.0333, P<0.05) between the average DTF values under the condition with no music and classical music; a significant difference (P=0.0061, P<0.05) between the mean DTF values without music and rock music; no significant difference between the mean DTF values with classical music and rock music. In beta band, there is, no significant difference between the mean DTF values without music and classical music; a significant difference (P=0.0082, P<0.05) between the average DTF values without music and rock music; no significant difference between the average DTF values with classical music and rock music.

Table 1.

Average DTF values under different conditions $(\bar{x} \pm S)$ $(\bar{x}\pm S)$

	No music	Classical music	Rock music
Alpha band	0.13386±0.00179	0.10161±0.00063	0.08943±0.00079
Beta band	0.13860±0.00079	0.11982±0.00031	0.10513±0.00064

The threshold T=0.13 is used to binarize the DTF matrixes. The final DTF are as follows:

With these DTF matrixes, the whole brain network topologies of different band under different conditions are constructed as follows:

It shows that in alpha band and beta band, the number of network connection without music is significantly more than that under the condition with classical music and rock music and the two-way connections are especially obvious. Compared with those without music, the network connections of frontal area, central area and top area are reduced under the condition with classical music and rock music, and some connections change from two-way connection to one-way connection. Some one-way information flows change their directions or become two-way. In the frontal region, the network connections under the condition with classical music are more than those under the condition with rock music. While in the temporal lobe, the situation is opposite.

5

Analysis of network characteristics

In order to better describe the topology of the network, we analyzed some network topology parameters. We selects parameters such as node degree, global efficiency, connection density and information flow gain to study and analyze the constructed network[21-24].

5.1

Degree

The average node degrees of alpha band and beta band under different conditions are as follows:

It shows that in alpha band and beta band, the average node degrees with classical music and rock music are reduced compared with those without music and the average degrees with rock music are lowest, which indicates that classical music and rock music cause the decline of the strength of information transmission and rock music has more significant effect. T-test is used to explore the differences between the mean degrees of the same band under different conditions. The result shows that in alpha band, there is, a significant difference (P=0.0260, P<0.05) between the average degrees under the condition with no music and classical music; a significant difference (P=0.0237, P<0.05) between the average degrees under the condition with no music and rock music; no significant difference between the mean degrees under the condition with classical music and rock music. In beta band, there is, no significant difference between the average degrees under the condition with no music and classical music; a significant difference (P=0.0111, P<0.05) between the average degrees under the condition with no music and rock music; no significant difference between the average degrees with classical music and rock music.

In order to analyze the information flow of the brain network under threes conditions, we counted the five nodes with the most output information and the five nodes with the most inflow information of the alpha band and beta band under three conditions. The results are as follows:

Table 2.

Five nodes with the most output or inflow information in alpha and beta band brain networks under different conditions

No.	No music				Classical music				Rock music
	alpha		beta		alpha		beta		alpha		beta
	out	in	out	in	out	in	out	in	out	in	out	in
1	F8	T8	FC1	O1	F8	O1	F8	T8	F8	T7	FZ	T7
2	F7	FC6	F8	FC6	O2	FC2	C6	T7	FC1	O2	FC1	O1
3	P2	T7	F4	P8	F7	F4	FZ	O1	C1	TP7	F4	O2
4	AF4	FC5	FZ	T8	C6	T7	FC1	FC5	CP1	O1	CP2	TP7
5	P1	C1	P1	TP8	FC6	C1	AF3	CP6	FZ	P7	F3	T8

points are located in the frontal region and parietal region, and the main information inflow points are located in the central region; under the condition with classical music, the main information outflow points are located in the frontal region and parietal region, some nodes have more information output and the main information inflow points are located in the parietal region; under the condition with rock music, the main information outflow points are located in the frontal region and central region and the main information inflow points are located in the parietal region. In beta band, without music the main information outflow points are located in the frontal region and the node that has the most information output is FC1 and the main information inflow points are located in the parietal region and central region and most of them are in the right brain; under the condition with classical music, the main information outflow points are located in the frontal region and central region, and the main information inflow points are located in the temporal lobes on the left and right sides of the central region; under the condition with rock music, the main information outflow points are located in the frontal region and the main information inflow points are located in the parietal region and the sides of the central region.

The changes of the brain regions for information inflow and outflow indicate that background music activated some nodes in the central region and parietal region during the working memory tasks and increased the strength of information transmission. Meanwhile, it restrained the excitability of some nodes in the frontal region, which caused the decline of the strength of information transmission between these nodes. In general, the negative effect of background music on brain network is greater than the positive effect.

5.2

Global efficiency

The global efficiencies of alpha band and beta band under three conditions are as follows:

As is shown in the figures, the global efficiency of alpha band and beta band with music is lower than that without music, which indicates that the speeds of information transmission between network nodes get lower with music stimulation. T-test is used on them, which shows that in alpha band, there is, a significant difference (P=0.0331, P<0.05) between the global efficiencies with no music and classical music; a significant difference (P=0.0170, P<0.05) between the global efficiencies with no music and rock music; no difference between the global efficiencies with classical music and rock music. In beta band, there is, no difference between the global efficiencies with no music and classical music; a significant difference (P=0.0082, P<0.05) between the global efficiencies with no music and rock music; no difference between the global efficiencies with classical music and rock music.

5.3

Connection density

The connection densities of alpha band and beta band under three conditions are as follows:

The connection densities of alpha band with classical music and rock music are lower than that without music, which indicates that the node connections get less and the connection tightness between nodes is weakened. T-test is used on them, which shows that in alpha band, there is, no difference between the connection densities under the condition with no music and classical music; a significant difference (P=0.0016, P<0.05) between the connection densities under the condition with no music and rock music; a significant difference (P=0.0198, P<0.05) between the connection densities under the condition with classical music and rock music. In beta band, there is, no significant difference between three conditions.

5.4

Information flow gain

The information flow gains of alpha band and beta band under three conditions are as follows:

As is shown in the figure, the main brain active areas are in the frontal region and central region. The information flow gain of some nodes in the frontal region under the condition with rock music is much higher than those under the condition with no music and classical music, which indicates that the rock music stimulation makes this region more active and this region contributes more to the causal network.

6

Conclusions

We concluded that working memory is related to the activities of frontal and parietal regions. Classical music and rock music reduce the causal connection strength between nodes and the efficiencies of information transfer between brain regions. Music stimulations reduce the correct response rates and increase the response time.

The conclusions indicate that background music is redundant for the working memory and interferes with the activities of the related brain regions, which causes the increase of the brain load and the decline of the working memory.

Język:: Angielski

Częstotliwość wydawania:: 1 razy w roku
Dziedziny czasopisma:: Nauki biologiczne, Nauki biologiczne, inne, Matematyka, Matematyka stosowana, Matematyka ogólna, Fizyka, Fizyka, inne

Kanał RSS czasopisma

Research on the Effect of Background Music on Working Memory Based on Granger Causal Network

Lingyue Wang

Lei Guo

Xinsheng Yang

Ying Li

Data publikacji: 27 lut 2025

Otrzymano: 23 wrz 2024

Przyjęty: 09 sty 2025

DOI: https://doi.org/10.2478/amns-2025-0137

Słowa kluczoweEEG, Working Memory, Background Music, Granger Causal Network

© 2025 Lingyue Wang et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Słowa kluczowe
EEG, Working Memory, Background Music, Granger Causal Network