Reproduce “MixAssist: An Audio-Language Dataset for Co-Creative AI Assistance in Music Mixing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Michael Clemens, Ana Marasović
A new audio-language dataset, MIXASSIST, supports co-creative AI assistance in music mixing by capturing multi-turn, audio-grounded dialogues between experts and amateurs. Fine-tuning state-of-the-art Audio-Language Models on this dataset enables them to provide contextually relevant mixing advice, with human evaluations showing a slight preference for AI-generated responses in some aspects.
Key insights
- Fine-tuning ALMs on a domain-specific, audio-grounded conversational dataset like MIXASSIST is crucial for generating helpful and contextually relevant mixing advice.
- Human users desire AI as an explainable teacher and technical assistant that provides visually integrated, controllable feedback, while expressing concerns about creative homogenization.
- Current ALMs still demonstrate limitations in deep, accurate audio analysis and reasoning specific to nuanced music mixing contexts, despite improvements from fine-tuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.