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

Xia, Xianjun; Togneri, Roberto; Sohel, Ferdous; Huang, Defeng

doi:10.1016/j.patcog.2018.03.025

Cited by 45 publications

(18 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Değer et al [15] proposed a classification algorithm based on RF, which solved the problem of poor robustness in face emotion recognition. Xiao et al [38] greatly improved the success rate of acoustic event monitoring system by using RF classifier. Fabris et al [17] used RF to predict gene expression as the brain ages.…”

Section: Introductionmentioning

confidence: 99%

Four-classes Human Emotion Recognition Via Entropy Characteristic and Random Forest。

Zhang

Min²

2020

ITC

View full text Add to dashboard Cite

Human Emotion Recognition is of vital importance to realize human-computer interaction (HCI), while multichannel electroencephalogram (EEG) signals gradually replace other physiological signals and become the main basis of emotional recognition research with the development of brain-computer interface (BCI). However, the accuracy of emotional classification based on EEG signals under video stimulation is not stable, which may be related to the characteristics of EEG signals before receiving stimulation. In this study, we extract the change of Differential Entropy (DE) before and after stimulation based on wavelet packet transform (WPT) to identify individual emotional state. Using the EEG emotion database DEAP, we divide the experimental EEG data in the database equally into 15 sets and extract their differential entropy on the basis of WPT. Then we calculate value of DE change of each separated EEG signal set. Finally, we divide the emotion into four categories in the two-dimensional valence-arousal emotional space by combining it with the integrated algorithm, Random Forest (RF). The simulation results show that the WPT-RF model established by this method greatly improves the recognition rate of EEG signal, with an average classification accuracy of 87.3%. In addition, we use WPT-RF model to train individual subjects, and the classification accuracy reached 97.7%.

show abstract

Section: Introductionmentioning

confidence: 99%

Four-classes Human Emotion Recognition Via Entropy Characteristic and Random Forest。

Zhang

Min²

2020

ITC

View full text Add to dashboard Cite

show abstract

“…Corresponding author: h.phan@kent.ac.uk reduction [22,7]. Particularly, the multitasking approach that jointly performs event detection and event boundary estimation [23,6,24] has demonstrated state-of-the-art performance on different benchmark datasets. In the latter stream, overlapping events are either separated using source separation methods [25,26] prior to detection, or recognized via a selection of local spectral features [11,10,7].…”

Section: Introductionmentioning

confidence: 99%

“…However, there exists a methodological gap between them. Audio events intrinsically possess temporal structures, and tailoring a network's output layer and loss functions for structure modelling has been shown to be efficient for the isolated AED [23,6,24]. However, this capacity has been uncharted for overlapping AED, and it remains questionable how to generalize a network's output layer and tailor its loss functions [20,6] to accommodate arbitrary event overlap, i.e.…”

Section: Introductionmentioning

confidence: 99%

Unifying Isolated and Overlapping Audio Event Detection with Multi-label Multi-task Convolutional Recurrent Neural Networks

Phan

Chén

Koch

et al. 2019

ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

We propose a multi-label multi-task framework based on a convolutional recurrent neural network to unify detection of isolated and overlapping audio events. The framework leverages the power of convolutional recurrent neural network architectures; convolutional layers learn effective features over which higher recurrent layers perform sequential modelling. Furthermore, the output layer is designed to handle arbitrary degrees of event overlap. At each time step in the recurrent output sequence, an output triple is dedicated to each event category of interest to jointly model event occurrence and temporal boundaries. That is, the network jointly determines whether an event of this category occurs, and when it occurs, by estimating onset and offset positions at each recurrent time step. We then introduce three sequential losses for network training: multi-label classification loss, distance estimation loss, and confidence loss. We demonstrate good generalization on two datasets: ITC-Irst for isolated audio event detection, and TUT-SED-Synthetic-2016 for overlapping audio event detection.

show abstract

“…SVM have been used with the energy and the signal spectrogram in [18] and [19] for surveillance application, and also in [20] respectively, using a deep architecture (DCNN). Other authors considered also Random Forest [24] using more contextual information.…”

Section: Introductionmentioning

confidence: 99%

Real-Time monophonic and polyphonic audio classification from power spectra

Baelde

Biernacki

Greff³

2019

Pattern Recognition

View full text Add to dashboard Cite

This work addresses the recurring challenge of real-time monophonic and polyphonic audio source classification. The whole normalized power spectrum (NPS) is directly involved in the proposed process, avoiding complex and hazardous traditional feature extraction. It is also a natural candidate for polyphonic events thanks to its additive property in such cases. The classification task is performed through a nonparametric kernel-based generative modeling of the power spectrum. Advantage of this model is twofold: it is almost hypothesis free and it allows to straightforwardly obtain the maximum a posteriori classification rule of online signals. Moreover it makes use of the monophonic dataset to build the polyphonic one. Then, to reach the real-time target, the complexity of the method can be tuned by using a standard hierarchical clustering preprocessing of the prototypes, revealing a particularly efficient computation time and classification accuracy trade-off. The proposed method, called RARE (for Real-time Audio Recognition Engine) reveals encouraging results both in monophonic and polyphonic classification tasks on benchmark and owned datasets, including also the targeted real-time situation. In particular, this method benefits from several advantages compared to the state-of-the-art methods including a reduced training time, no feature extraction, the ability to control the computation -accuracy trade-off and no training on already mixed sounds for polyphonic classification.

show abstract

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

Cited by 45 publications

References 40 publications

Four-classes Human Emotion Recognition Via Entropy Characteristic and Random Forest。

Four-classes Human Emotion Recognition Via Entropy Characteristic and Random Forest。

Unifying Isolated and Overlapping Audio Event Detection with Multi-label Multi-task Convolutional Recurrent Neural Networks

Real-Time monophonic and polyphonic audio classification from power spectra

Contact Info

Product

Resources

About