Multi-label Cross-lingual automatic music genre classification from lyrics with Sentence BERT

TL;DR

This work tackles cross-lingual, multi-label music genre classification from lyrics, focusing on English and Portuguese and eight overlapping genres. It uses multilingual Sentence-BERT embeddings (paraphrase-multilingual-mpnet-base-v2) to obtain song-level representations by averaging sentence embeddings, followed by a one-vs-all SVM classifier for multi-label prediction; a TF-IDF Bag-of-Words baseline with translation is used for comparison. Compared to the translation-based baseline, the sBERT-based approach substantially improves genrewise F1 from around $0.35$ to about $0.69$ on cross-lingual tasks, and dataset centralization further boosts cross-lingual performance. This work offers a scalable path for genre classification across underrepresented languages and highlights residual cultural biases and domain-shift challenges in cross-lingual music information retrieval.

Abstract

Music genres are shaped by both the stylistic features of songs and the cultural preferences of artists' audiences. Automatic classification of music genres using lyrics can be useful in several applications such as recommendation systems, playlist creation, and library organization. We present a multi-label, cross-lingual genre classification system based on multilingual sentence embeddings generated by sBERT. Using a bilingual Portuguese-English dataset with eight overlapping genres, we demonstrate the system's ability to train on lyrics in one language and predict genres in another. Our approach outperforms the baseline approach of translating lyrics and using a bag-of-words representation, improving the genrewise average F1-Score from 0.35 to 0.69. The classifier uses a one-vs-all architecture, enabling it to assign multiple genre labels to a single lyric. Experimental results reveal that dataset centralization notably improves cross-lingual performance. This approach offers a scalable solution for genre classification across underrepresented languages and cultural domains, advancing the capabilities of music information retrieval systems.

Figures (1)

• Figure 1: Proposed method. Lyrics are first processed by a pre-trained sBERT. This yields embeddings which are further classified by a Support Vector Machine, whose output is a genre prediction. A Bag-of-Words representation based on a TF-IDF vectorizer is used as a baseline for performance comparison.