Tapping Into a New Paradigm: A Synthetic Strategy for Automatic Drum TapScription

André Santos; Amílcar Cardoso; Matthew E. P. Davies; Roger B. Dannenberg

Tapping Into a New Paradigm: A Synthetic Strategy for Automatic Drum TapScription

Abstract:

We introduce Automatic Drum TapScription (ADTS), a novel paradigm for rhythmic interaction consisting of transcribing arbitrarily-timbred taps into drum representations. Our approach targets taps produced on a variety of surfaces without other controlled timbral characteristics other than playing style. Our long-term goal is to enable more accessible and creative percussive exploration but presents significant challenges due to the minimal timbre variation between taps intended to represent different drum classes. To address these challenges, we take the first steps toward achieving ADTS by designing an effective dataset synthesis strategy. This strategy enables new opportuneties for musical expression by considering drumming at a more semantic or functional level as opposed to a simple collection of timbres. We present initial results, comparing three different models: one trained on drum data, another trained on a small dataset of quasi-aligned tapped performances, and another trained on our synthetic dataset. Our synthetic approach shows promise, demonstrating progress in this untapped domain.