SPHERE: Automatic Music Upmixing via Audio Language Model Post-Training with Spatial Heuristic Rewards
2026-08-31 • Sound
Sound
AI summaryⓘ
The authors explore how to automatically adjust the spatial sound in music mixes using a special type of audio language model, rather than building a custom system for the task. They train the model further with a two-step process: first by rejecting bad examples, then by using smart feedback based on clear rules derived from music mixing practices. They introduce a set of rules called Sphere that help the model make mixes that sound balanced and well-placed in space. Their work shows that expert knowledge about music mixing can be turned into measurable rewards to teach language models without needing new designs.
automatic music upmixingaudio language model (ALM)post-trainingrejection samplingreinforcement learningverifiable rewardsGRPOmusic mixing conventionsspatial sound mixing
Authors
Zixun Guo, Calvin Murdock, Sanjeel Parekh, W Owen Brimijoin, Simon Dixon, Joshua Reiss, Ishwarya Ananthabhotla
Abstract
In this paper, we study the task of automatic music upmixing, wherein a system predicts spatial mixing parameters from a multi-stem recording. Different from existing methods that rely on task-specific music encoders, we approach this task via audio language model (ALM) post-training, leveraging rich representations from existing ALMs, which encode both music semantics and mixing knowledge. Specifically, we propose a post-training recipe that first employs rejection sampling SFT, followed by reinforcement learning (RL) with verifiable rewards (RLVR) via GRPO. We propose Sphere (Spatial Heuristic Rewards), a deterministic reward suite inspired by music mixing conventions, to guide our post-training. It consists of 6 perceptually-motivated sub-rewards and encourages the output mix to be centered, balanced and spacious. More broadly, our results suggest that expert domain knowledge can be encoded as verifiable rewards and distilled into language models, without task-specific architectures.