Personalized Music Improvisation

This research paper focuses on developing a model that learns and emulates the personalized musical style of composers. Pitch estimation of music signals is a fundamental problem in automatic transcription systems, with applications in music information retrieval and automated musicological analysis. While pitch estimation for monophonic music signals is considered solved, estimating the pitch of multiple concurrent sources remains challenging. This paper explores the challenges and techniques involved in monophonic pitch detection, highlighting the differences from polyphonic detection and emphasizing the need for improved accuracy. Additionally, the paper discusses the potential of using machine learning algorithms to synchronize music in real-time with a composer's input, enabling interactive improvisation. The project's scope includes the development of algorithms for pitch detection, synchronization, and AI-based melody generation, with applications ranging from music composition assistance to interactive performance systems.