| dc.description.abstract | The imitation of percussive sounds via the human voice is a natural and effective tool for communicating rhythmic ideas on the fly. Query by Vocal Percussion (QVP) is a subfield in Music Information Retrieval (MIR) that explores techniques to query percussive sounds using vocal imitations as input, usually plosive consonant sounds. In this way, fully automated QVP systems can help artists prototype drum patterns in a comfortable and quick way, smoothing the creative workflow as a result. This project explores the potential usefulness of recent data-driven neural network models in two of the most important tasks in QVP. Algorithms relative to Vocal Percussion Transcription (VPT) detect and classify vocal percussion sound events in a beatbox-like performance so to trigger individual drum samples. Algorithms relative to Drum Sample Retrieval by Vocalisation (DSRV) use input vocal imitations to pick appropriate drum samples from a sound library via timbral similarity. Our experiments with several kinds of data-driven deep neural networks suggest that these achieve better results in both VPT and DSRV compared to traditional data-informed approaches based on heuristic audio features. We also find that these networks, when paired with strong regularisation techniques, can still outperform data-informed approaches when data is scarce. Finally, we gather several insights relative to people’s approach to vocal percussion and how user-based algorithms are essential to better model individual differences in vocalisation styles. | en_US |
