My name is Jadon Nguyen! I am currently a second year student at UCLA studying computer engineering. Recently, I have been fascinated by the concept of using artificial intelligence for image, video, and audio generation. These ideas hold many implications for the future of modern media. While the concept of AI generated media is not new, the concept of more personalized applications of these processes is yet to be fully explored. The project at hand focuses on using electrodermal activity data in order to measure one’s physiological response to specific music in order to generate new samples of music catered towards the user. The objective of this is to create content that feeds the user’s subconscious reactions and foster a profound connection with the AI-generated music, resulting in an immersive and personalized experience. If this application of the technology was further developed years from now, it is possible that it would greatly benefit areas of music therapy, cultural preservation, and storytelling.
If this application of the technology was further developed years from now, it is possible that it would greatly benefit areas of music therapy, cultural preservation, and storytelling.
The components of this experiment can be dissected into three distinct elements: the electrodermal activity sensor, the groove generation, and the learning algorithm. The electrodermal activity (EDA) measures the electrical conductivity of the skin. This conductivity is affected by sweat glands, which changes based on the autonomic nervous system response. The EDA provides information vectors about one’s reaction to specific stimuli. In this case, the greater the value indicates a stronger reaction to the provided groove. After recording such a response, the information is provided to a reinforcement learning algorithm which will generate the parameters for the next groove sample. For example, if one exhibits a greater response to a groove that is more jazz-influenced, the algorithm will provide the necessary inputs to develop a jazz-centric groove. There are many programs and methods available to generate the sound file. However, the ideal program is able to generate relatively random files, influenced upon a given user input. The current program in consideration is a very novel project titled “Los Angeles Music Composer.” The documentation can be found here. It is capable of continuous generation of music, with varying levels of control for each component (instruments, randomness, duration, etc.).
The current stage of the project is primarily focused on understanding the different components at hand. This means ensuring that the EDA equipment exhibits the correct behavior, with the data able to be correctly parsed, and the AI music generation able to consistently produce a variety of outputs. If a year’s worth of time and data are put into this project, I envision that the scope of this project can expand to full symphonic pieces generated based on user emotions. Ideally, with improved hardware and more developed models, visual generation components and real-time feedback are feasible goals.