Transfer Learning for Violinist Identification

Yudong Zhao, György Fazekas, Mark Sandler

Abstract: Music performer identification is important for music recommendation, music expression analysis and playlist generation. In previous research, audio feature learning methods were commonly used for both singer identification (SID) and instrument player identification (IPID) with good results. In the current deep learning era, SID results are greatly improved using neural networks, however, instrument player identification is rarely investigated in recent works primarily due to the shortage of open-access datasets. To solve this problem, we construct a concerto violin dataset as well as a solo dataset, and present a transfer learning approach for violinist identification from pre-trained music auto-tagging neural networks and singer identification models. We then transfer pre-trained weights and fine-tune the models using violin datasets and finally obtain violinist identification results. We compare our system with a number of state-of-the-art methods and show that our model outperforms them using both of our datasets.

External IDs:dblp:conf/eusipco/ZhaoFS22