MuLan: A Joint Embedding of Music Audio and Natural Language

TL;DR

MuLan is a novel joint embedding model that links music audio directly to natural language descriptions, enabling versatile zero-shot music tagging and cross-modal retrieval across diverse genres and text styles.

Contribution

Introduces MuLan, a joint audio-text embedding model trained on 44 million recordings, allowing flexible, zero-shot music understanding beyond traditional ontology-based systems.

Findings

Effective zero-shot music tagging demonstrated

Versatile cross-modal retrieval capabilities shown

Supports diverse music genres and natural language descriptions

Abstract

Music tagging and content-based retrieval systems have traditionally been constructed using pre-defined ontologies covering a rigid set of music attributes or text queries. This paper presents MuLan: a first attempt at a new generation of acoustic models that link music audio directly to unconstrained natural language music descriptions. MuLan takes the form of a two-tower, joint audio-text embedding model trained using 44 million music recordings (370K hours) and weakly-associated, free-form text annotations. Through its compatibility with a wide range of music genres and text styles (including conventional music tags), the resulting audio-text representation subsumes existing ontologies while graduating to true zero-shot functionalities. We demonstrate the versatility of the MuLan embeddings with a range of experiments including transfer learning, zero-shot music tagging, language understanding in the music domain, and cross-modal retrieval applications.