Drew Edwards

Analysis of robustness and overfitting in piano transcription systems; introduces a re‑performed MAESTRO dataset and evaluates data augmentation approaches to improve out-of-distribution performance; includes a brief state-of-the-art (SOTA) evaluation on the MAPS piano transcription dataset.

Paper on domain adaptation methods for high-resolution guitar transcription (co-authored with QMUL collaborators, including Riley et al.), contributing to state-of-the-art guitar transcription results.

Research on converting audio/MIDI to guitar tablature using masked language models; part of broader transcription and symbolic modelling work.

Introduces PiJAMA, a large dataset of jazz piano performances with automatic MIDI annotations; includes dataset analysis and benchmark experiments for performer identification and generative modeling.

Co-inventor on patents related to object/identity recognition and cross-source image object identification from work in applied ML/computer vision contexts (listed on Google Scholar/patent summaries).