Acoustic Event Classification using spectral band selection and Non-Negative Matrix Factorization-based features

Ludeña-Choez, Jimmy; Gallardo-Antolín, Ascensión

doi:10.1016/j.eswa.2015.10.018

Cited by 21 publications

(6 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Like typical automatic classification systems, most of the approaches for environmental sound classification rely on handcrafted features or learn representations from mid-level representations such as spectro-temporal features (Ludeña-Choez & Gallardo-Antolín, 2016;Costa et al, 2012). Spectral representations have been used as features in several approaches based on matrix factorization (Mesaros et al, 2015;Benetos et al, 2016;Bisot et al, 2016;Salamon & Bello, 2015;Geiger & Helwani, 2015).…”

Section: Introductionmentioning

confidence: 99%

End-to-end environmental sound classification using a 1D convolutional neural network

Abdoli

Cardinal

Koerich

2019

Expert Systems with Applications

257

128

View full text Add to dashboard Cite

In this paper, we present an end-to-end approach for environmental sound classification based on a 1D Convolution Neural Network (CNN) that learns a representation directly from the audio signal. Several convolutional layers are used to capture the signal's fine time structure and learn diverse filters that are relevant to the classification task. The proposed approach can deal with audio signals of any length as it splits the signal into overlapped frames using a sliding window. Different architectures considering several input sizes are evaluated, including the initialization of the first convolutional layer with a Gammatone filterbank that models the human auditory filter response in the cochlea. The performance of the proposed end-to-end approach in classifying environmental sounds was assessed on the UrbanSound8k dataset and the experimental results have shown that it achieves 89% of mean accuracy. Therefore, the propose approach outperforms most of the state-of-the-art approaches that use handcrafted features or 2D representations as input. Furthermore, the proposed approach has a small number of parameters compared to other architectures found in the literature, which reduces the amount of data required for training.

show abstract

Section: Introductionmentioning

confidence: 99%

End-to-end environmental sound classification using a 1D convolutional neural network

Abdoli

Cardinal

Koerich

2019

Expert Systems with Applications

257

128

View full text Add to dashboard Cite

show abstract

“…In fact, the selected features should capture the important characteristics of different kinds of the acoustic signals. The feature selection processes can be categorized as wrapper methods (model-dependent) and filter techniques (modelindependent) [64]. In this paper, the wrapper methods have been chosen, particularly in the form of neighbourhood component analysis (NCA).…”

Section: Feature Selectionmentioning

confidence: 99%

Acoustic signal processing with robust machine learning algorithm for improved monitoring of particulate solid materials in a gas flowline

Aminu

McGlinchey

Cowell

2019

Flow Measurement and Instrumentation

View full text Add to dashboard Cite

Acoustic signal processing with robust machine learning algorithm for improved monitoring of particulate solid materials in a gas pipelineThe flow of particulate solid materials in a gas pipeline can significantly erode mechanical equipment, and hence, real-time quantitative monitoring is a timely need for the oil and gas industry. Although a considerable amount of research has been conducted employing acoustic signals for qualitative monitoring, there is still an unmet demand for a simple and robust quantitative monitoring system. Acoustic signal processing with machine learning is a simple and robust method that has the potential to meet this demand, but has not been previously used for particulate solid material quantitative monitoring. Here we report on the development of acoustic signal processing methods strictly on the existence and the significance of the correlation between emitted acoustic signals and the flow conditions and behaviours of particle-laden gas pipeline. The integrated, conventional Artificial Neural Network (ANN) models are used to capture the distribution of the acoustic feature vectors extracted from the signal processing techniques. The backpropagation learning method coupled with Grey wolf optimiser is used to adjust the weights of the network to minimize the regularized cost function for each feature vector. The Grey wolf optimiser is used to provide global adaptation strategy for the network hyper-parameters. The results from the signal processing techniques demonstrate a significant qualitative association between flow conditions and the emitted acoustic signature. Further, conventional ANN has mainly been concerned with capturing systematic patterns in a distribution of measurements fixed in time and the results of the processes are collected in discrete time intervals. Therefore, a modification of the classical ANN, called the Time Delay Neural Networks (TDNN) is used to capture such dynamics. The proposed method compares the performance of the classical ANN models with the TDNN models wherein the feature vectors were used to train the TDNN models.Results show that the TDNN models outperform the classical ANN models which confirm the fact that classical ANN models are insufficient for processing these time sequences. Overall, this study lays the basis for employing signal processing techniques in the development of a real-time quantitative particulate solid monitoring in a gas pipeline.

show abstract

“…Authors in [38] investigated the use of biologically-inspired features, derived from a filter-bank of two-dimensional Gabor functions. In [39], spectral band selection based features are used.…”

Section: Related Workmentioning

confidence: 99%

Random forest classification based acoustic event detection utilizing contextual-information and bottleneck features

Xia

Togneri

Sohel

et al. 2018

Pattern Recognition

View full text Add to dashboard Cite

The variety of event categories and event boundary information have resulted in limited success for acoustic event detection systems. To deal with this, we propose to utilize the long contextual information, low-dimensional discriminant global bottleneck features and category-specific bottleneck features. By concatenating several adjacent frames together, the use of contextual information makes it easier to cope with acoustic signals with long duration. Global and category-specific bottleneck features can extract the prior knowledge of the event category and boundary, which is ideally matched by the task of an event detection system. Evaluations on the UPC-TALP and ITC-IRST databases of highly variable acoustic events demonstrate the effectiveness of the proposed approaches by achieving a 5.30% and 4.44% absolute error rate improvement respectively compared to the state of art technique.

show abstract

Acoustic Event Classification using spectral band selection and Non-Negative Matrix Factorization-based features

Abstract: Acoustic event classification using spectral band selection and non-negative matrix factorization-based features. Expert Systems with Applications, 46, 77-86.

Cited by 21 publications

References 32 publications

End-to-end environmental sound classification using a 1D convolutional neural network

End-to-end environmental sound classification using a 1D convolutional neural network

Acoustic signal processing with robust machine learning algorithm for improved monitoring of particulate solid materials in a gas flowline

Random forest classification based acoustic event detection utilizing contextual-information and bottleneck features

Contact Info

Product

Resources

About