ISMIR Paper: Unifying Drum Transcription and Stem Generation via Latent Diffusion
affige_yang · x · 2026-08-05
Automatic Drum Transcription (ADT) typically maps a music mixture directly to symbolic events, discarding useful acoustic stems for editing.
This upcoming ISMIR paper introduces a Separate-and-Detect architecture:
- Five-stem Latent Diffusion: Jointly generates kick, snare, toms, hi-hats, and cymbals in a compact VAE latent space.
- Fixed Onset Detector: Converts separated stems into symbolic events.
- Training-only Branches: Uses an Onset branch (OB) and a Timbre branch (TB) to shape the separator during training, which are discarded at inference.
Evaluated on MDB Drums and ENST-Drums, the pipeline outperforms a strong U-Net separation baseline in overall transcription F1. It also beats an end-to-end ADT system on kick and snare F1 while providing editable separated audio stems.
More from Research
- Prior Labs Acquired by SAP; TabPFN Creator on Tabular Data — ziv_ravid · 2026-08-26
- Dataset of 115,293 illustrated pages from Encyclopaedia Britannica (1768-1929) released on Hugging Face — vanstriendaniel · 2026-08-26
- Study: GLM 5.2 shows consistent performance across different APIs — niloofar_mire · 2026-08-26
- Anthropic's Jack Lindsey to Discuss Claude's J-Space and Consciousness in Webinar — PeterBowdenLive · 2026-08-26
- Study finds Agent harness impacts benchmark scores more than the model itself — rohanpaul_ai · 2026-08-26
- NeurIPS 2026 Workshop Aims to Ground LLM Interpretability as Rigorous Science — ninamiolane · 2026-08-26