International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 10 Issue: 08 | Aug 2023
p-ISSN: 2395-0072
www.irjet.net
Music Engendering Algorithm With Diverse Characteristic By Adjusting The Parameters Of Deep Convolutional Generative Adversarial Networks Aitik Dandapat1, Dr. Aishwarya Dandpat2 1Undergraduate Student, Veer Surendra Sai University of Technology, Burla 2DNB Resident, Ramkrishna Care Hospital
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - With the widespread development of deep
information. The piano roll (a picture-like representation of data) was used to encode information, with one axis representing tempo and the other axis representing pitch. Each pixel intensity in the range of 0–128 was represented by the note's velocity. DCGAN was chosen as the deep learning architecture used in this article. It can capture and learn the data distribution from a data set and generate a sample from the same distribution. Finally, the model generates sample snippets: some are synthesized with musical dynamics and some are not. These extracts were mixed with normal music for analysis and a user study was conducted.
learning, automatic composition has become a highly topical topic occupying the minds of music computer scientists. The paper proposes advanced arithmetic for music engenderation using Generative Adversarial Networks (GANs). The music is divided into tracks and the note segment of the tracks is expressed as a piano roll by a trained GAN model whose generator and discriminator play a continuous zero-sum game to produce high musical integrity. In most cases, although GAN excels in image generation, the model adopts a cross-channel deep convolutional network structure in accordance with the properties of the music data in this article, producing music more closely matched to human hearing and aesthetics.
2. RELATED WORK Algorithmic composition is a subject of work that dates back to 1959. However, the recent developments of Deep Neural Networks (DNN), which have proven astonishing results in learning from big datasets, allowed this topic of music generation to be further developed. Over the past couple of years, tons of proposed models addressing music generation have been published, all of them on deep learning algorithms.
Key Words: Generative Adversarial Network, DGCAN, Convolutional layer, MusPy, Pianoroll
1. INTRODUCTION Music Engenderation composes music on the computer using machine learning methods. With the increase in AI data and research, researchers have started exploring machine learning in music creation.
2.1 Recurrent Neural Network The most popular architecture in music creation systems is the recurrent neural network (RNN). It is a feedforward neural network, meaning that the output from the hidden layers is transmitted back to it and the layer below it. As a result, it has the capacity to record information and pick up knowledge from sequence data. However, because the weight matrix is updated continuously during backpropagation, RNN experiences the vanishing and exploding gradient problem.
Deep learning is a branch of machine learning methods that can recognize patterns and make decisions without explicit programming. It shows promising results in several areas such as natural language processing for texts, computer vision for images, and speech recognition for speech. In addition, deep learning techniques in artificial music engenderation have successfully generated human-like compositions. However, most research has focused on musical composition and neglected the aspect of expressive musical performance. Therefore, musical information stored on MIDI (Musical Instrument Digital Interface) tracks, such as speed, proved useless during practice. This makes the music produced rather mechanical and boring.
Long Short-Term Memory (LSTM), an RNN variant, uses a variety of gates to overcome the issue in a unique RNN. These gates include input, output, forget, and cell state gates. Information can be sent through cell states as a conduit. An applicant is chosen by the input gates. The input gates pick a candidate from the inputs and change of any previously available pertinent data. The forget gates, on the other hand, purge the cell states of unimportant data. The information that is sent to the next concealed state is finally decided by the output gates.
This document concerns the design of a music production system that could produce music with built-in speed, also known as musical dynamics. To do this, the training data must encode speed as well as altitude and time
© 2023, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 617