Convolutional Neural Network with Multi-Task Learning Scheme for Acoustic Scene Classification

Convolutional Neural Network with Multi-Task Learning Scheme for Acoustic Scene Classification
Title:
Convolutional Neural Network with Multi-Task Learning Scheme for Acoustic Scene Classification
Other Titles:
APSIPA ASC 2017
DOI:
Publication URL:
Keywords:
Publication Date:
01 December 2017
Citation:
Abstract:
Deep Neural Network (DNN) with Multi-Task Learning (MTL) methods have recently demonstrated significant performance gains on a number of classification, detection, recognition tasks compared to conventional DNN. DNN with MTL framework involves cross-task and within-task knowledge sharing layers. MTL methods have benefit for regularization effect from the cross-task knowledge sharing layers. And, within-task knowledge sharing layers allow MTL based DNN to learn information to optimize the performance for individual task. We formulate our acoustic scene classification in MTL framework using Convolutional Neural Network to learn information specific to different types of environment. We conduct experiments using DCASE2016 dataset. Proposed approach achieves 83.8% accuracy to classify 15 acoustic scene classes.
License type:
PublisherCopyrights
Funding Info:
Description:
ISBN:

Files uploaded:

File Size Format Action
proof-apsipa17-dcase2017-v11-cr-ieee1-1.pdf 97.48 KB PDF Open