AUTHOR=Yang LiDong , Yue RenBo , Wang Jing , Liu Min 

TITLE=Neural Network Model Based on the Tensor Network for Audio Tagging of Domestic Activities

JOURNAL=Frontiers in Physics

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2022.863291

DOI=10.3389/fphy.2022.863291

ISSN=2296-424X

ABSTRACT=Due to the serious problem of population aging, monitoring of domestic activities is increasingly important. Audio tagging of domestic activities is very suitable when the visual data is unavailable due to interference from light and the environment. Aiming at solving this problem, a neural network model based on tensor network is proposed for audio tagging of domestic activities that is more interpretable than traditional neural networks. The introduction of tensor network can compress the network parameters and reduce the redundancy of the training model while maintains good performance. Firstly, important features of Mel spectrogram of the input audio are extracted through CNN (Convolutional Neural Networks). Then they are converted into the high-order space corresponding with tensor network. The spatial structure information and important features can be further extracted and retained through MPS (matrix product state). Large patches of feature data are divided into small local orderless patches when using tensor network. Final tagging results are obtained through MPS layers which is just a tensor network structure based on tensor train decomposition. In order to evaluate the proposed method, the DCASE2018 challenge task5 dataset for monitoring domestic activities is selected. The results show that the average F1-score of the proposed model in the test set of development dataset and validation dataset reach 87.7% and 85.9%, which are 3.2% and 2.8% higher than the baseline system respectively. It is verified that the proposed model can perform better and more efficiently for audio tagging of domestic activities.