Last week, the Atari reading group came to an end. This week, we will have a smooth transition into the second topic: music and AI. We will read RL-Duet: Online Music Accompaniment Generation Using Deep Reinforcement Learning. This is an exiting paper fusing the realms of RL and Music by creating a virtual band mate improvising with you. Unlike traditional offline music generation, the algorithm dynamically responds to human input, generating a machine counterpart in a sequential order. The key innovation lies in our reinforcement learning approach, where the generation agent learns a policy based on a well-functioning reward model derived from monophonic and polyphonic training data. Preliminary experiments demonstrate the algorithm’s ability to seamlessly adapt to human input, producing melodic, harmonic, and diverse machine-generated accompaniments of superior quality compared to baseline methods.
Home