Interactive VST Plugin composing tool for generating short melodic lines using genetic algorithms. Plugin is desired to work as a MIDI Effect, which enables user to to quickly generate and modify melody ideas in any time signature / scale, using desired qualities like number of dissonances or chord conformance, as well as control parameters / automation of the instruments.
VST (Virtual Studio Technology) plugins are software components that can be integrated into digital audio workstations (DAWs) and other compatible software for music production. These plugins extend the functionality of the host software by providing additional effects, instruments, and processing capabilities.
There exist a fair amount of VST plugins designed to help musicians in their creative process. Some of them involve simple music theory tools [1], allowing to explore variety of chords and scales in creative process. Some modern plugins use advanced AI models for generation, allowing them to introduce innovative functionalities. There are a few positions on the market that are worth mentioning. There is a set of 5 AI tools from Google called Magenta Studios, which allow to completely transform existing melodies and drum patterns [2]. A plugin called Synplant allows to create synth patches from audio recordings provided by the user [3]. Another tool called Orb Producer Suite allows for chord / melody generation, quick randomization of patterns and their customization [4]. InstaComposer 2 [11] offers DAW-like interface for creating multiple tracks in different modes using chosen scales. WavTool [12] has similar approach, but uses LLM in creative process to help choosing presets or scales.
Music generation models based on neural networks are fairly effective, but their limitations are lack of controllability and restriction to dominant musical structures, like 4/4 common time signature [5]. Alternative approach might be using genetic algorithms, which are based simply on optimization of the melody, rather than predicting consecutive notes based on learned data. The biggest challenge for building a genetic algorithm is creating a fitness function to evaluate quality of the created melodies. One approach is user evaluating melodies based on their preference, as presented in widely cited GenJam [6] work. However, using that approach is very time consuming, as it requires user evaluating all members of population for every generation. For many years research has been conducted concerning finding appropriate fitness function. Many of works use music theory rules as a source of evaluation. [7,9] measure qualities such as amount of dissonances or scale conformance in the melody, when [7,8] use statistical criteria to find the distance between qualities measured in generated melody and the same qualities in real pieces. [5] used that approach to evaluate melodies according to desired mood, playing style and composition parameters, which are predicted from text using LLM. There are also another approaches worth analysing, such as using sequence models like LSTM as a fitness function for genetic algorithm [10].
To our best of knowledge, there is not yet a VST plugin that uses genetic algorithms for the task of MIDI generation, at least not in a way we proposed. We also did not find an approach that concerns generating rhythm and melody seperately, which we find to have a great potential in creative process.
[1] Scaler 2. https://www.scalerplugin.com/. Access: 10.03.2024\ [2] Magenta Studio. https://magenta.tensorflow.org/studio/. Access: 10.03.2024\ [3] Synplant. https://soniccharge.com/synplant/. Access: 11.03.2024\ [4] Orb Producer Studio 3. https://www.orbplugins.com/orb-producer-suite/. Access: 11.03.2024\ [5] Poćwiardowski J., (2023). System for synthetizing music from users description [Unpublished bachelor’s thesis]. Gdańsk University of Technology.\ [6] Biles J. “GenJam: A genetic algorithm for generating jazz solos”. in: ICMC. T. 94. Ann Arbor, MI. 1994, p. 131–137.\ [7] Zdzisław Kowalczuk, Marek Tatara & Adam Bąk. “Evolutionary music composition system with statistically modeled criteria”. in: Trends in Advanced Intelligent Control, Optimization and Automation: Proceedings of KKA 2017—The 19th Polish Control Conference, Kraków, Poland, June 18–21, 2017. Springer. 2017, p. 722–733.\ [8] L Rocha de Azevedo Santos, Carlos Nascimento Silla Jr & MD Costa-Abreu. “A methodology for procedural piano music composition with mood templates using genetic algorithms”. in: (2021).\ [9] Michael Towsey “Towards melodic extension using genetic algorithms”. in: Educational Technology and Society (2001), p. 54–65.\ [10] Farzaneh, Majid, and Rahil Mahdian Toroghi. "GGA-MG: Generative genetic algorithm for music generation." arXiv preprint arXiv:2004.04687 (2020).\ [11] InstaComposer 2. https://www.waproduction.com/plugins/view/instacomposer-2. Access 12.03.2024\ [12] WavTool. https://wavtool.com/. Access 12.03.2024