Automatic Drum Transcription

My PhD in Audio Engineering and Deep Learning at Birmingham City University focused on automatic drum transcription: detecting and classifying drum events directly from audio recordings.

The research explored recurrent and convolutional neural networks, soft attention mechanisms, multiple time-step loss functions for better peak picking, and a "player vs transcriber" game approach to data manipulation. It also produced a review of the field published in IEEE/ACM Transactions on Audio, Speech, and Language Processing.

Alongside the papers I built ADTWeb, an open source browser-based drum transcription system, contributed to MDB Drums (an annotated subset of MedleyDB), and submitted systems to the MIREX 2017 drum transcription evaluation.