CREATING A SOFTWARE APPLICATION FOR GENERATING MUSICAL COMPOSITIONS USING A NEURAL NETWORK MODEL
Authors:
1.
Kazan National Research Technological University
2.
Kazan National Research Technological University
Type:
Article
Pages:
from 86
to 90
Status:
Published
Received:
24.12.2025
Accepted:
25.12.2025
Published:
25.12.2025
Language:
Russian
Keywords:
NEURAL NETWORK MODEL LSTM, GENERATION OF MUSICAL COMPOSITIONS, MIDI COMPOSITIONS, TOKENIZATION OF MUSICAL DATA
Abstract (English):
The presented work is devoted to the description of the creation of a software application for the automatic generation of MIDI musical compositions using the LSTM neural network model. Various neural network architectures are analyzed: LSTM, GAN, transformers and diffusion models. The LSTM model was chosen as the most stable and adaptive for generating sequences, trained on musical fragments, followed by generating melodies and saving the results in MIDI format. Tokenization of music data (in MIDI format) was implemented through the pianoroll representation, which made it possible to transform music into sequences suitable for submission to a neural network. Implemented neural network architecture and user interface. A user interface has been developed that allows users to set generation parameters and download the finished file, as well as a server logic system on Flask. Testing was carried out to confirm the operability of the system and the quality of the resulting musical sequences. During the development of the application, problems related to the performance of the model and the quality of music generation were also identified and eliminated. The result is a working prototype that automatically generates music. Main product characteristics: generation of MIDI compositions with parameter settings (tempo, length, creative variability); simple and intuitive interface local data processing that does not require an Internet connection; free prototype at the development stage, with the prospect of switching to a freemium model.
The presented work is devoted to the description of the creation of a software application for the automatic generation of MIDI musical compositions using the LSTM neural network model. Various neural network architectures are analyzed: LSTM, GAN, transformers and diffusion models. The LSTM model was chosen as the most stable and adaptive for generating sequences, trained on musical fragments, followed by generating melodies and saving the results in MIDI format. Tokenization of music data (in MIDI format) was implemented through the pianoroll representation, which made it possible to transform music into sequences suitable for submission to a neural network. Implemented neural network architecture and user interface. A user interface has been developed that allows users to set generation parameters and download the finished file, as well as a server logic system on Flask. Testing was carried out to confirm the operability of the system and the quality of the resulting musical sequences. During the development of the application, problems related to the performance of the model and the quality of music generation were also identified and eliminated. The result is a working prototype that automatically generates music. Main product characteristics: generation of MIDI compositions with parameter settings (tempo, length, creative variability); simple and intuitive interface local data processing that does not require an Internet connection; free prototype at the development stage, with the prospect of switching to a freemium model.
Keywords:
NEURAL NETWORK MODEL LSTM, GENERATION OF MUSICAL COMPOSITIONS, MIDI COMPOSITIONS, TOKENIZATION OF MUSICAL DATA
NEURAL NETWORK MODEL LSTM, GENERATION OF MUSICAL COMPOSITIONS, MIDI COMPOSITIONS, TOKENIZATION OF MUSICAL DATA
