Development of gunfire sound classification system with a smartphone using ANN

Tangkawanit, Settha; Pinthong, Chairat; Kanprachar, Surachet

doi:10.1109/icdamt.2018.8376517

Cited by 8 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is seen that this figure can be divided into 3 main parts; that is, the input signal preparation part (to make the captured gunfire sound into the digital format), the feature extraction part (to prepare the feature vector for ANN), and the classification part (to determine the gun type). The feature extraction and the classification process are identical to that from [1][2]. The major task in this research is to study about techniques in selecting the feature vector as the input to the feature extraction process.…”

Section: Proposed Classification Systemmentioning

confidence: 99%

“…As mentioned previously, the feature extraction in this work is the one used in [1][2]; that is, the Frequency Spectral Representative Bin (FSRB) is adopted. The diagram of this part is also shown in Fig.…”

Section: Feature Extractionmentioning

confidence: 99%

“…The ANN model used in this work is shown in Fig. 9 [1]. It consists of 4 layers; that is, Input Layer, Hidden Layer 1, Hidden Layer 2, and Output Layer, respectively.…”

Section: Classificationmentioning

confidence: 99%

“…It is seen that these spectral-based techniques can also be applied to the gunfire analysis since both are considered as impulses. This spectral consideration has been done also in gunfire sound classification as shown in [1][2]. The frequency components of the gunfire sound in the frequency range of interest were used as feature vectors in an ANN model for performing the classification process.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Effect of Sampling Rate Reduction and Signal Filtering for Gunfire Sound Classification with Spectral Vector using ANN

Tangkawanit

Kanprachar

2020

ECTI-EEC

Self Cite

View full text Add to dashboard Cite

To identify the type of the shooting gun, the frequency components of the gunfire sound has been utilized as the input to an Artificial Neural Network (ANN) model [1-2] with a promising recognition accuracy. With this method, the required computational resource is large if the sampling frequency is too high. In this research, the sampling frequency for transforming the input gunfire sound into a digital signal is varied from 44.1 kHz down to 4.41 kHz. 6 different types of gunfire sound are considered. Additionally, the effect of applying different types of signal filtering is studied. It is found that only reducing the sampling frequency on the input gunfire sound signal does not deliver a good performance on gunfire sound classification. To obtain a good classification accuracy, signal filtering has to also be applied. With Chebyshev Type II filter and 4.41 kHz sampling frequency, the classification accuracy is all 100% for the practical range of SNR. This impressive classification accuracy comes with a 10-times reduction on the computational resource; that is, from the sampling frequency of 44.1 kHz to 4.41 kHz. The findings from this work can be applied to the gunfire sound classification system with limited computational resource.

show abstract

Section: Proposed Classification Systemmentioning

confidence: 99%

Section: Feature Extractionmentioning

confidence: 99%

“…The ANN model used in this work is shown in Fig. 9 [1]. It consists of 4 layers; that is, Input Layer, Hidden Layer 1, Hidden Layer 2, and Output Layer, respectively.…”

Section: Classificationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Effect of Sampling Rate Reduction and Signal Filtering for Gunfire Sound Classification with Spectral Vector using ANN

Tangkawanit

Kanprachar

2020

ECTI-EEC

Self Cite

View full text Add to dashboard Cite

show abstract

“…Note that these closed commercial systems are not available for test. Other works classify emergency related sounds leveraging machine learning [42], [81], [98] using different set of features and models, while others [48], [98], [109] use Neural Networks (NNs) [48], [98], [109] for classifying the captured audio. Notably, for the home scenario, glass break detection capabilities are employed as it can be evidenced by Amazon manufactured devices such as Alexa [24], and Google devices such as the nest hub [38] and nest mini [39].…”

Section: B Gunshot and Suspicious Sound Detectionmentioning

confidence: 99%

Audio Attacks and Defenses against AED Systems -- A Practical Study

Santos¹,

Nilizadeh²

2021

Preprint

View full text Add to dashboard Cite

Audio Event Detection (AED) Systems capture audio from the environment and employ some deep learning algorithms for detecting the presence of a specific sound of interest. In this paper, we evaluate deep learning-based AED systems against evasion attacks through adversarial examples. We run multiple security critical AED tasks, implemented as CNNs classifiers, and then generate audio adversarial examples using two different types of noise, namely background and white noise, that can be used by the adversary to evade detection. We also examine the robustness of existing third-party AED capable devices, such as Nest devices manufactured by Google, which run their own black-box deep learning models.We show that an adversary can focus on audio adversarial inputs to cause AED systems to misclassify, similarly to what has been previously done by works focusing on adversarial examples from the image domain. We then, seek to improve classifiers' robustness through countermeasures to the attacks. We employ adversarial training and a custom denoising technique. We show that these countermeasures, when applied to audio input, can be successful, either in isolation or in combination, generating relevant increases of nearly fifty percent in the performance of the classifiers when these are under attack.

show abstract