The goal of this work is to present an audio signal classification system based on Linear Predictive Coding and Random Forests. We consider the problem multiclass with imbalanced datasets. signals under belong class sounds from wildlife intruder detection applications: birds, gunshots, chainsaws, human voice tractors. proposed achieves overall correct rate 99.25%. There no probability false alarms in case birds or voices. For other three classes low, around 0.3%. omission also low: 0.2% for...

This paper presents a grid search approach to optimize the kernel's parameters for support vector machines classifier. The most encountered three kernels are considered: linear, radial basis, and sigmoid. We show that optimization of improves recognition performance audio signals classification, especially in case sigmoid kernel. behavior model is very sensitive parameters, which, turn data selected features. consider problem multiclass classification with imbalanced datasets. compare...

Walking is the most basic form of human activity for achieving mobility. As an essential function body, examination walking directed towards assessment body mechanics in posture and during movement. This work proposes a wearable smart system monitoring objective evaluation foot biomechanics gait. The proposed solution assumes cross-correlation plantar pressure with lower-limb muscular activity, throughout stance phase walking. Plantar acquired array resistive sensors deployed onto shoe...

Service robots are extensively used these days for industry and assisted living applications. In this work we present how the capabilities of TIAGo robot can be extended. We show that it handle tasks received through audio signals. Experimental results provided to support our claims.

The goal of this paper is to present three modules that can be used independently or in conjunction for audio signal processing purposes. Here, the are illustrated context acoustic analysis TIAGo service robot (recording data stream, obtaining isolated events, generating classification model, identifying events). For two them friendly user interfaces described. signals under belong a home environment. As features we propose liftering Mel Frequency Cepstral Coefficients, while...

Supercapacitors are components for energy storage, dedicated applications where both and power density needed. Even if their is ten times lower than the of batteries, supercapacitors offer new alternatives storage We have done an experimental bench we measured voltage across a supercapacitor in constant current charge/discharge cycle. Using these measurements mathematical algorithm proposed electrical model supercapacitor, determined parameters also.

In this paper, we study several audio classification schemes applied on different number of features for multiclass with imbalanced datasets. As features, proposed the liftering Mel frequency cepstral coefficients, while use probabilistic methods, instance-based learning algorithms, support vector machines, neural networks, L <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">∞</sup> -norm based classifier, fuzzy lattice reasoning and trees. The...

In this paper we shall present recent results of two applications for monitoring using acoustical signal classification. The first case study is the problem context awareness based on acoustic analysis a service robot. Then discussed classification wildlife intruder detection. Previous are briefly recalled and new experimental also provided.

The synergy between batteries and capacitors has been growing, to the point where supercapacitos may soon be almost as indispensable portable electricity are now. One very important parameter of a supercapacitor is Equivalent Series Resistance (ESR). energy that can stored in depends on ESR, so we intend measure parameter. Our tests were made 350F supercapacitor, 14 V, produced by Econd Ltd. We have two methods: first consist evaluating ESR constant current discharge, second method consists...

In this work we compare different classification algorithms applied on number of features (linear predictive coding coefficients) in order to detect audio signals from wildlife areas. The final goal is find the appropriate linear coefficients provide desired accuracy for a certain framework. experimental results prove that best classifier Logistic Model Trees regardless features, having constant greater than 95%. case reduced both Random Forest and Lazy IBk have good results; 98%.

The task of creating a short, accurate and fluent summary starting from larger text document or group documents is called summarization. When the generated by extracting parts original text, process known as extractive This work focused on use convex optimization positive defined symmetric kernels for summarization text. paper includes two new contributions. First, we show how kernel method can be used unsupervised Second, investigate empirically performance different functions with respect...

The paper describes a convex optimization formulation of the extractive text summarization problem and simple scalable algorithm to solve it. program is constructed as relaxation an intuitive but computationally hard integer programming problem. objective function highly symmetric, being invariant under unitary transformations representations. Another key idea replace constraint on number sentences in summary with surrogate. For solving we have designed specific projected gradient descent...

Many zones, especially wild areas need surveillance systems in order to protect them against potential destroying actions. An attractive solution for automatic consists audio based systems, which present some advantages compared with video or mixt systems. This paper proposes an alarming events detection system two levels. On the first level detects only if event is a dangerous one (alarm on) normal off). second level, it case, identifies exactly nature of four classes: chainsaw, gunshot,...

In this paper some properties of spectrograms and sparsograms are reviewed. The framework addressed is acoustic based wildlife intruder detection. spectral signatures also recalled within framework. averaged binary sparsogram introduced it shown that can be considered an effective tool for classifying the possible intruders sounds into different classes.

The goal of this work is to present some possible intruder detection systems and the influence impulse-like signals upon overall classification accuracy. Two different scenarios are used: in first scenario five sound classes considered (last class belong impulsive sounds - gunshots), while second we dropped out class. More classifiers used both number features considered. An improvement accuracy obtained within scenario. highest for J48 classifier using 51 features, attained Simple Logistic...

This paper presents a way in which TIAGo, service robot from Pal Robotics, could be integrated everyday life, based on his audio capabilities. We show that using 20 Mel frequency cepstral coefficients and 5-nearest neighbors we rich an overall correct classification rate of 99.27% the testing phase. However, do not recommend non-MFCC features for implementations with TIAGo robot, at least framework presented current paper.

The goal of this paper is to perform a comparison on different classification algorithms applied Mel-Frequency Cepstral Coefficients and Moving Picture Experts Group-7 features in order obtain high average correct rate, greater than 98%, low computation time, less 1 minute, for audio purposes the case signals from service robots. highest rates are obtained using Linear Discriminant Analysis phase. For averaged accuracy 99.78%, 64 features, time 5.26 seconds. 99,65%, with 5.30

In this work we study the quantization of audio signals for a zero-crossing method recently used to detect intruders in wildlife areas. This implements two descriptors: D (represents number samples between real zeros) and S points local minima/maxima consecutive zeros). We show using experimental results that proposed based intruder detection framework, D/S pairs are almost constant till bits is less than six.

Facial recognition applications present a great interest in the area of computer vision, with various methods and approaches that provide impressive performance. However, not all studies investigate possibilities using proper feature extraction efficient classifiers, for facial expression is required detection. In this sense, we propose another application based on Local Binary Patterns or fusion Discrete Cosines Transform extraction, classifier Simplified Fuzzy Adaptive Resonance Theory Map...

In this paper we propose a new algorithm for extractive text summarization. The proposed method is based on convex minimization and the properties of l1 norm. has some advantages, like extensibility ability to easy take into account additional information constraints. It provides very good results, but its execution time usually larger than that other similar procedures.

In this paper some results on positive trigonometric polynomials related to one-dimensional discrete phase retrieval problem are presented. The closed forms of the z-transform and Fourier transform folded circular autocorrelation provided show that whether scaling or biasing input magnitude data should have an effect non-negativity. Some enlighten relationship between nearest method also discussed.

Abstract In this paper we present an updated version of the audio database acquired by TIAGo service robot and simulated robot. To initial which consists in 1380 signals have added 1920 more acoustic signals. The now 3300 isolated events corresponding to 110 classes. All recorded sound correspond indoor environment, they are spread into five different scenarios: kitchen, room, appliances, voice non-verbal. is intended be used order identify based on signature, especially when elderly or...

Having a piecewise linear fitting of gain, one can obtain good approximation phase. There are few ways to determine the slopes broken line approximation, but most popular approaches need breakpoints. In this paper we breakpoints using divide and conquer approach. Results gain resulting phase provided for different cases.