Artigos publicados
Artigo 1: Determining a Non-Intrusive Voice Quality Model Using Machine Learning and Signal Analysis in Time
Abstract. Telecommunications services monitoring is of paramount importance, network operators must meet the expectations of their subscribers. On wireless communications, transmitted signal may suffer different degradations in the transmission channel. In the literature, there are different models of wireless channels, among the most cited are the Rician and Rayleigh fading channel models, each one presents different characteristics of transmission. In the speech quality assessment research area, most of the works focused on wired network type degradations, using parameters as packet losses, delay and jitter; however, wireless degradation characteristics are not considered. In this context, the present paper investigate the impact of different Doppler frequency shifts (Hz) on speech communications, using the fading channel models previously mentioned, and different multipath delay profiles. Performance evaluation parameters are based on the voice quality assessment algorithms described in recommendations ITU-T P.563 and P.862. The experimental results show how the variation range of the Doppler shift is related to speech quality index.
https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/572
Artigo 2: Determining a Non-Intrusive Voice Quality Model Using Machine Learning and Signal Analysis in Time
Abstract. The purpose of this paper is to determine a solution to estimate the quality of a signal of using time domain signal information and machine learning algorithms in an environment that simulates wireless networks using Voice over Internet Protocol (VoIP). The methodology employed was divided into three stages, and degradations were initially applied in an environment that simulated wireless networks making changes in two parameters being the signal-to-noise ratio (SNR) and the type of modulation scheme. To perform the degradations on six distinct signals, algorithms implemented in MATLAB were used to simulate the effect of fading in wireless environments. In the second step, time domain graphs were plotted that correspond to the degradations and that were saved, 272 of them were used for training on 12 different learning algorithms. implemented in the Weka tool. In the last step, software-trained algorithms implemented in Java called PredictorFX in order to predict the value of MOS through an audio image in the time domain. The results were satisfactory, the best trained regression algorithms called r1 were RandomTree, RandomForest and IBk with correlation coefficients ranging from 0.9798 to 0.9982 in the validation phase. In relation to r2 the best were RandomTree, RandomForest, IBk and AditiveRegression with correlation coefficient ranging from 0.9375 to 0.9923 in the validation phase. And finally, for the training algorithms for the named c1 classification the best trained algorithms were IBk, RandomTree, RandomForest and J48 with a range of 48.53% to 98.53% of correctly classified instances.
https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/630
Artigo 3: Ensemble of hybrid Bayesian networks for predicting the AMEn of broiler feedstuffs
To adequately meet the nutritional needs of broilers, it is necessary to know the values of apparent metabolizable energy corrected by the nitrogen balance (AMEn) of the feedstuffs. To determine AMEn values, biological assays, feedstuff composition tables, or prediction equations are used as a function of the chemical composition of feedstuffs, the latter using statistical methodologies such as multiple linear regression, neural networks, and Bayesian networks (BN). BN is a statistical and computational methodology that consists of graphical (graph) and probabilistic models of quantitative and/or qualitative variables. Ensembles of BN in the area of broiler nutrition are expected, as there is no research showing their AMEn prediction performance. The purpose of this article is to propose and use ensembles of hybrid Bayesian networks (EHBNs) and obtain prediction equations for the AMEn of feedstuffs used in broiler nutrition from their chemical compositions. We trained 100, 1,000, and 10,000 EHBN, and in this way, empirical distributions were found for the coefficients of the covariates (crude protein, ether extract, mineral matter, and crude fiber). Thus, the mean, median, and mode of these distributions were calculated to build prediction equations for AMEn. It is observed that the method for obtaining the coefficients of the covariates discussed in this article is an unprecedented proposal in the field of broiler nutrition. The data used to obtain the equations were obtained by meta-analysis, and the data for the validation of the equations were obtained from metabolic tests. The proposed equations were evaluated by precision measures such as the mean square error (MSE), mean absolute deviation (MAD), and mean absolute percentage error (MAPE). The best equations for predicting AMEn were derived from the mean or mode coefficients for the 10,000 EHBN results. In conclusion, the methodology used is a good tool to obtain prediction equations for AMEn as a function of the chemical composition of their feedstuffs. The coefficients were found to differ from those found by other methodologies, such as the usual neural network or multiple linear regressions. The field of broiler nutrition contributed with new equations and with a never-applied methodology and differentiated in obtaining its coefficients by empirical distributions.