Environmental Sound Classification Based on Multi-temporal Resolution Convolutional Neural Network Combining with Multi-level Features

Abstract

Motivated by the fact that characteristics of different sound classes are highly diverse in different temporal scales and hierarchical levels, a novel deep convolutional neural network (CNN) architecture is proposed for the environmental sound classification task. This network architecture takes raw waveforms as input, and a set of separated parallel CNNs are utilized with different convolutional filter sizes and strides, in order to learn feature representations with multi-temporal resolutions. On the other hand, the proposed architecture also aggregates hierarchical features from multi-level CNN layers for classification using direct connections between convolutional layers, which is beyond the typical single-level CNN features employed by the majority of previous studies. This network architecture also improves the flow of information and avoids vanishing gradient problem. The combination of multi-level features boosts the classification performance significantly. Comparative experiments are conducted on two datasets: the environmental sound classification dataset (ESC-50), and DCASE 2017 audio scene classification dataset. Results demonstrate that the proposed method is highly effective in the classification tasks by employing multi-temporal resolution and multi-level features, and it outperforms the previous methods which only account for single-level features.

Environment Sound Classification using Multiple Feature Channels and Attention based Deep Convolutional Neural Network

“…The best results were produced when Gammatone cepstral coefficients were combined TEO-GTSC with score-level fusion. A multi-temporal resolution CNN was proposed in [34]. Here, multiple CNNs with different filter sizes and stride lengths work on a raw audio signal on different temporal resolutions, in parallel.…”

An End-to-End Audio Classification System Based on Raw Waveforms and Mix-Training Strategy

“…[12] proposed a timedomain network based on the 1D raw waveform. The results in [13,14] demonstrated that the feature directly learned from the raw waveform is sufficient for audio classification problems. However, these works are designed specially for single-label audio classification.…”

Fluid Temperature Detection Based on its Sound with a Deep Learning Approach

“…Considering the sound processing field, few studies have been conducted in this field, especially classifying and grading fluids sounds, except the analysis and recognition of human voices, which is a highly active area and works such as speech [34][35][36][37][38][39][40] recognition have been performed in this regard. However, studies have been recently conducted on the classification of fluids sounds in other areas, including human heartbeat [41], urban sounds [42][43][44][45][46][47][48] play sounds, car horns, air conditioning sound, engine sounds, etc. ), and music [49][50][51][52][53][54][55][56], using deep learning.…”