Adding a consistency loss between vocal-only and mixture encoder features during LoRA fine-tuning of Whisper improves lyrics transcription on music mixtures by a small but consistent margin, without needing source separation at inference.
The authors would like to thank Vin´ıcius Marques, Angelo Maugeri, Ana Rachel Melo Nascimento, and Ana Lu ´ısa Runze from Music.AI for their help with data annotation and cleaning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Enhancing Lyrics Transcription on Music Mixtures with Consistency Loss
Adding a consistency loss between vocal-only and mixture encoder features during LoRA fine-tuning of Whisper improves lyrics transcription on music mixtures by a small but consistent margin, without needing source separation at inference.