Home >

news Help

Publication Information

English:Closed-Form Pre-Training for Small-Sample Environmental Sound Recognition 
Japanese: 井上中順, Goto Keita.  
English: Nakamasa Inoue, Keita Goto.  
Language English 
Journal/Book name
English:2020 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) 
Volume, Number, Page         pp. 1693-1697
Published date Dec. 31, 2020 
Conference name
English:Asia-Pacific Signal and Information Processing Association Annual Summit and Conference 2020(APSIPA ASC) 
Conference site
Official URL http://www.apsipa.org/proceedings/2020/APSIPA-ASC-2020.html
Abstract This paper presents a framework for pre-training neural networks, namely closed-form pre-training, and we apply it to small-sample environmental sound recognition. Our main idea is to pre-train neural networks on a dataset automatically gener- ated by some formulas, without any prior real-world recordings or manual annotation. Specifically, the proposed framework consists of two steps. First, an audio classification dataset is generated. Here, we propose three types of dataset definitions using colored noise and its extensions. Second, a network is pre-trained on the generated dataset. The obtained pre-trained network is particularly effective for fine-tuning with few examples because it helps optimization methods avoid falling into a premature local optimal solution. In experiments, we demonstrate the effectiveness of the proposed framework for small-sample environmental sound recognition on three datasets: ESC-10/50, and UrbanSound8K. We obtained performance improvement on all datasets with a small number of training samples.

©2007 Tokyo Institute of Technology All rights reserved.