Year: 2024
Type: article
Abstract: Automatic Drum-set Transcription (ADT) aims to convert drum performance audio into corresponding musical notes. Unlike ordinary instruments, drum performances are characterized by higher discreteness, faster tempos, and shorter note durations. To address these challenges, we propose a novel method for achieving precise drum-set music transcription. Our approach employs a Transformer model as the feature extractor and applies the SemiCRF loss function to guide the prediction probabilities of all potential notes. Given the scarcity of drum-set transcription datasets within the community, we have collected and curated a high-quality, detailed-labeled dataset of drum performances spanning various styles and rhythms, totaling over 1000 minutes. Comparative experimental results demonstrate the efficacy of our proposed method.
Author: Shijie Hao
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4405592503