Search papers, labs, and topics across Lattice.
This paper introduces MMGenre, a novel benchmark designed to evaluate singing voice synthesis (SVS) across ten major musical genres and 26 subgenres, addressing the limitations of existing benchmarks that predominantly focus on pop music. The study reveals that current SVS models exhibit weak genre discrimination, producing similar acoustic characteristics across different genres, with only marginal improvements from zero-shot genre adaptation. However, the authors find that lightweight genre-specific continued training can significantly enhance performance, highlighting the need for more nuanced approaches in genre-aware SVS development.
Current singing voice synthesis models struggle with genre discrimination, showing that genre-specific training can dramatically improve performance where zero-shot methods fail.
Singing voice synthesis (SVS) has progressed rapidly, yet its ability to generalize across diverse musical genres remains underexplored. Existing benchmarks are heavily biased toward pop music, limiting systematic analysis of genre-dependent behavior. We introduce MMGenre, a benchmark for multi-genre SVS diagnosis, supported by an automatic pipeline for constructing genre-aligned music scores. MMGenre spans 10 major genres and 26 subgenres, enabling comprehensive analysis of genre-aware synthesis. Extensive evaluation of representative SVS models reveals limited genre discrimination: synthesized vocals across genres exhibit highly similar acoustic characteristics and weak separability. While zero-shot genre adaptation yields only marginal improvements, lightweight genre-specific continued training leads to substantial gains. MMGenre provides a standardized framework for multi-genre SVS evaluation and exposes critical challenges in achieving genre-aware singing voice synthesis.
