Life Tips & Miscellaneous Travel and History Sports and Arts Zen and Life Tips Mathematics and Machine Learning Clojure Java Books, TV, Movies and Music Navigation of this blog
Rhythm and melody and the autonomic nervous system
As discussed in ‘Emotions, the autonomic nervous system and the “regulating” effect’, the human autonomic nervous system is greatly influenced by emotions, and the autonomic nervous system is influenced by the external environment.
The rhythms and melodies of music have attracted attention in music therapy, psychology and neuroscience research as factors that influence the autonomic nervous system. For example, the rhythm of music affects the physiological rhythm of the body (heartbeat and breathing), and when the body moves to a certain rhythm, it resonates with the rhythm and the heart rate may stabilise; melodies evoke emotions and affect the mind, with bright melodies causing positive emotions and vice versa sad melodies evoke negative emotions. Calm and pleasant melodies also stimulate the parasympathetic nervous system and promote relaxation and peace of mind.
Because of these effects, music therapy is widely used as a means of regulating the autonomic nervous system and reducing stress and anxiety, using rhythm and melody to regulate clients’ emotional and physical responses and promote mental health.
Recent research has also shown that musical rhythms and melodies can influence the secretion of neurotransmitters and hormones in the brain, and in particular, the secretion of dopamine and serotonin is regulated by music.
The relationship between sound, rhythm and melody and UX
Thus, sound, rhythm and melody have a strong influence on people’s emotions and are very important elements in the user experience (UX).
For example, with regard to sound, sound feedback to actions, such as the sound of a button click or notification sound, can make it easier for users to recognise an action, calm music can make them feel relaxed, upbeat music can make them feel lively, certain sounds and jingles can increase brand awareness and are used, for example.
In terms of rhythm, the rhythm of the user’s actions can affect the smoothness and intuitiveness of the interface, rhythmic feedback can make it easier for the user to continue with an action, the right rhythm can increase concentration while working and provide an environment in which the user is immersed in the task, and The user can also be immersed in the task.
Furthermore, with regard to melodies, pleasant melodies can help users to recall the app or website, memorable melodies can have the effect of encouraging return visits, and the use of melodies can tell an emotional story and gain user empathy.
Sound, rhythm and melody are powerful tools for improving UX, and the conscious and effective use of these elements in UX design can help users feel comfortable and more connected to the brand.
Algorithm of pleasant sound
Let us now consider the algorithms that are used to make users feel comfortable. They incorporate principles of music theory and psychoacoustics, as shown below.
1. selection of scales and chords:
– Major and minor scales: major scales are brighter and minor scales are quieter. It is important to select the appropriate scale for the purpose of the music.
– Chord combinations: generally pleasing chord combinations (e.g. C-G-Am-F) can be used in chord progressions to create easy-listening songs.
2. rhythm and tempo:
– Balanced rhythms: suitably simple rhythmic patterns provide comfort to the listener. If the rhythm is too complex, it can be confusing.
– Appropriate tempo: a pleasant musical tempo, usually between 60 and 120 BPM (beats per minute), is considered to have a relaxing effect.
3. sound characteristics:
– Colour of sound: The waveform of a sound (e.g. sine, square or sawtooth waves) can give a different impression of the sound. Sine waves have a clear, pleasant sound, while sawtooth waves have a richer timbre.
– Volume dynamics: The smoother the change in volume, the easier it is to hear. Abrupt volume changes can tire the listener.
4. use of environmental sounds:
– Sampling of natural sounds: the inclusion of natural sounds such as birdsong, wind, running water, etc. can create a relaxing atmosphere.
– White and pink noise: adding these sounds to the background can amplify the pleasantness of the sound.
5. machine learning algorithms:
– Generative models: generative models such as GANs (Generative Adversarial Networks) and VAEs (Variational Autoencoders) can be used to learn pleasant sounds. This makes it possible to generate a wide variety of music styles.
– Feedback loops: algorithms can be implemented to adjust the style and tempo of music based on user feedback to provide a more personalised experience.
Examples of implementations of music generation using generative models
The following section outlines an example implementation for generating music using a general generative model that leads to the creation of pleasant sounds.
1. music generation using LSTM (long short-term memory) networks
LSTM, described in ‘Overview of LSTM, algorithms and implementation examples’, is a type of RNN suitable for generating time series data and can be used to generate musical melodies and rhythms.
The implementation procedure for LSTM is as follows:
1. preparing the dataset: collecting musical data, such as MIDI files, and converting them into a format that LSTM can train on.
2. building the model:
from keras.models import Sequential
from keras.layers import LSTM, Dense, Dropout
model = Sequential()
model.add(LSTM(128, input_shape=(timesteps, features), return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(128))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam')3. train the model: train the melody from previous musical data.
4. running the generation: after training, a sequence of random notes is generated.
2. music generation using GANs (Generation Against Networks)
GANs, described in ‘Overview of GANs and various applications and implementation examples’, are models in which generators and discriminators compete with each other to generate realistic data.
The implementation steps of a GAN are as follows:
1. preparing the dataset: collecting music data from MIDI and audio files
2. building the model:
from keras.models import Sequential
from keras.layers import Dense, Reshape, Flatten
generator = Sequential()
generator.add(Dense(256, input_dim=noise_dim))
generator.add(Reshape((16, 16)))
# Adding further layers.
discriminator = Sequential()
discriminator.add(Flatten(input_shape=(16, 16)))
discriminator.add(Dense(1, activation='sigmoid'))3. implementation of the training process: train the generator and the discriminator alternately.
4. performing the generation: after training, the noise vector is input to generate the music data.
3. music generation using VAE (Variational Auto Encoder)
The VAE, described in ‘Overview of the Variational Autoencoder (VAE) and examples of algorithms and implementations’, is a model for generating new data by learning potential representations of the data.
The implementation steps of a VAE are as follows:
1. preparing the dataset: preparing MIDI and audio files
2. building the model:
from keras.layers import Input, Dense, Lambda
from keras.models import Model
from keras import backend as K
input_shape = (timesteps, features)
inputs = Input(shape=input_shape)
x = Dense(128, activation='relu')(inputs)
z_mean = Dense(latent_dim)(x)
z_log_var = Dense(latent_dim)(x)3. training: train to minimise reconstruction error and KL divergence.
4. perform generation: generate new music data by sampling from the latent space.
4. using MusicVAE
MusicVAE, developed by Google’s Magenta project, is a specialised model for music generation.
The implementation steps using MusicVAE are as follows:
pip install magenta
2. model training: use pre-prepared models or train on your own data.
3. generation: generate new melodies and chords using trained models.
from magenta.models.music_vae import TrainedMusicVAE
music_vae = TrainedMusicVAE('path_to_trained_model')
generated_sequence = music_vae.sample(n=5) # Generates five melodies.Reference information and reference books
This section describes reference information and reference books on music generation using generative models. These books describe the theory of music generation, implementation methods and the fundamentals of deep learning.
4. Generating Music with a Generative Adversarial Network
5. Music Composition with Deep Learning: A Review
7. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow
コメント