Ignacio J. Turias

Improved likelihood ratio test based voice activity detector applied to speech recognition

Nowadays, the accuracy of speech processing systems is strongly affected by acoustic noise. This is a serious obstacle regarding the demands of modern applications. Therefore, these systems often need a noise reduction algorithm working in combination with a precise voice activity detector (VAD). The computation needed to achieve denoising and speech detection must not exceed the limitations imposed by real time speech processing systems. This paper presents a novel VAD for improving speech detection robustness in noisy environments and the performance of speech recognition systems in real time applications. The algorithm is based on a Multivariate Complex Gaussian (MCG) observation model and defines an optimal likelihood ratio test (LRT) involving multiple and correlated observations (MCO) based on a jointly Gaussian probability distribution (jGpdf) and a symmetric covariance matrix. The complete derivation of the jGpdf-LRT for the general case of a symmetric covariance matrix is shown in terms of the Cholesky decomposition which allows to efficiently compute the VAD decision rule. An extensive analysis of the proposed methodology for a low dimensional observation model demonstrates: (i) the improved robustness of the proposed approach by means of a clear reduction of the classification error as the number of observations is increased, and (ii) the trade-off between the number of observations and the detection performance. The proposed strategy is also compared to different VAD methods including the G.729, AMR and AFE standards, as well as other recently reported algorithms showing a sustained advantage in speech/non-speech detection accuracy and speech recognition performance using the AURORA databases.

Pitting corrosion behaviour of austenitic stainless steel using artificial intelligence techniques

Abstract Different artificial intelligent tools have been used to model pitting corrosion behaviour of EN 1.4404 austenitic stainless steel. Samples from this material have been subjected to polarization tests in different chloride solutions using different precursor salts: NaCl and MgCl2. The aim of this work is to compare the results obtained from the different classification models using both solutions studying the influence of them. Furthermore, in order to determine pitting potential values ( E pit ), different environmental conditions have been tested varying chloride ion concentration, pH value and temperature. The techniques used try to find the relation between the environmental parameters studied and the status pitting corrosion of this alloy. Several classification techniques have been used: Classification Trees (CT), Discriminant Analysis (DA), K-Nearest-Neighbours (K-NN), Back-Propagation Neural Networks (BPNN) and Support Vector Machine (SVM). The results obtained show the good correlation between experimental and predicted data for all the cases studied demonstrating the utility of artificial intelligence for modelling pitting corrosion problem.

A two-stage procedure for forecasting freight inspections at Border Inspection Posts using SOMs and support vector regression

The number of goods which passes through a border inspection post (BIP) may cause important congestion problems and delays in the port system, having an effect in the level of service of the port. Therefore, a prediction of the daily number of goods subject to inspection in BIPs seems to be a potential solution. This study proposes a two-stage procedure to better predict freight inspections. In the first stage, a Kohonen self-organising map (SOM) is employed to decompose the whole data into smaller regions which display similar statistical characteristics. In the second stage, support vector regression (SVR) is used to forecast the different homogeneous regions individually. The results obtained are compared with the single SVR technique. The experiment shows that SOM–SVR models outperform the single SVR models in the inspection forecasting. The application of the proposed technique may become a supporting tool for the prediction of the number of goods subject to inspection in BIPs of other international seaports or airports, and provides relevant information for decision-making and resource planning.

A Comparison of Forecasting Methods for Ro-Ro Traffic: A Case Study in the Strait of Gibraltar

The objective of this article is to predict volumes of Ro-Ro (Roll-on, Roll-off) freight in order to apply this prediction as a decision making tool in logistics planning and port organization. This tool can help to improve supply chain performance in a Ro-Ro terminal. Seasonal ARIMA (SARIMA) and Artificial Neural Networks (ANNs) were the forecasting methods used in this study. A resampling procedure was applied in order to find out the best model from a statistical point of view using multiple comparison methods. The results have been very promising (R=0.9157; d=0.9546; MSE=0.0195)

Breakdown potential modelling of austenitic stainless steel

The breakdown potential is a crucial factor in the study of pitting corrosion resistance of stainless steel. This work aims to demonstrate the advantage of different chemometric techniques to estimate the breakdown potential of austenitic stainless steel. In order to predict pitting corrosion behaviour of stainless steel, a total of 60 samples of this alloy were subjected to electrochemical tests varying chloride ion concentration, pH and temperature. The experimental values of the breakdown potential, in addition to the tested environmental factors, were used to construct the predictive models based on support vector machines and artificial neural networks. A multiple‐comparison study based on statistic tests was applied to determine the optimal configuration for each technique. According to the results, support vector machines became a suitable and reliable technique to be applied in the modelling of the breakdown potential of austenitic stainless steels. This technique outperformed the models based on artificial neural networks and provided a useful tool to compare the pitting corrosion resistance of stainless steel in different environmental conditions without recourse to polarization tests. Therefore, this model presented a relevant meaning in science and engineering applications. Copyright

An efficient VAD based on a generalized gaussian PDF

The emerging applications of wireless speech communication are demanding increasing levels of performance in noise adverse environments together with the design of high response rate speech processing systems. This is a serious obstacle to meet the demands of modern applications and therefore these systems often needs a noise reduction algorithm working in combination with a precise voice activity detector (VAD). This paper presents a new voice activity detector (VAD) for improving speech detection robustness in noisy environments and the performance of speech recognition systems. The algorithm defines an optimum likelihood ratio test (LRT) involving Multiple and correlated Observations (MCO). An analysis of the methodology for N = {2, 3} shows the robustness of the proposed approach by means of a clear reduction of the classification error as the number of observations is increased. The algorithm is also compared to different VAD methods including the G.729, AMR and AFE standards, as well as recently reported algorithms showing a sustained advantage in speech/non-speech detection accuracy and speech recognition performance.

A two‐stage forecasting approach for short‐term intermodal freight prediction

The forecasting of the freight transportation, especially the short-term case, is an important topic in the daily supply chain management. Intermodal freight transportation is subject to multiple complex calendar effects arising in the port environment. The use of prediction methods provides information that may be helpful as a decision-making tool in the management and planning of operations processes in ports. This work addresses the forecasting problem on a daily basis by a novel two-stage scheme combination to offer reliable predictions of fresh freight weight on Ro-Ro (roll-on/roll-off) transport for 7 and 14 days ahead. The study compares daily forecasting with a weekly forecasting approach. The applies database preprocessing and Bayesian regularization neural networks (BRNN) in Stage I. In Stage II, an ensemble framework of the best BRNN models is used to enhance the Stage I forecasting. The results show that the models assessed are a promising tool to predict freight time series for Ro-Ro transport.

Austenitic Stainless Steel EN 1.4404 Corrosion Detection Using Classification Techniques

Different methods of classification have been used in this paper to model pitting corrosion behaviour of austenitic stainless steel EN 1.4404. This material was subjected to electrochemical polarization tests in aqueous environment of varying chloride ion concentration (from NaCl solutions), pH values and temperature in order to determine values of critical pitting potentials (Epit) for each condition tested. In this way, the classification methods employed try to simulate the relation between Epit and those various environmental parameters studied. Different techniques have been used: Classification Trees (CT), Discriminant Analysis (DA), K-Nearest-Neighbours (K-NN), Backpropagation Neural Networks (BPNN) and Support Vector Machine (SVM). These models have generally been regarded as successful. They have been able to give a good correlation between experimental and predicted data. The analysis of the results becomes useful to plan improvement in the austenitic stainless steel protection and to avoid critical conditions expositures of this material.

Hybrid Approaches of Support Vector Regression and SARIMA Models to Forecast the Inspections Volume

The constant growth of air and maritime traffic of goods creates the need of increasing the number, the reliability and the security of inspections at the European borders. In this context of high security, this work applies a two-step procedure based on the hybridization of SARIMA and Support Vector Regression to forecast the inspection volume at the Border Inspection Post of Port of Algeciras Bay. Three hybrid approaches are proposed and two prediction horizons are evaluated. Based on several performance indexes to assess the goodness-of-fit of the models, the hybrid approaches perform better than the SARIMA and SVR models used separately. Hence, the study shows that the hybrid methodology improves the single methods. The experimental results can provide relevant information for resource planning and may become a decision-making tool in the inspection process of other European BIPs.

A constructive neural network to predict pitting corrosion status of stainless steel

The main consequences of corrosion are the costs derived from both the maintenance tasks as from the public safety protection. In this sense, artificial intelligence models are used to determine pitting corrosion behaviour of stainless steel. This work presents the C-MANTEC constructive neural network algorithm as an automatic system to determine the status pitting corrosion of that alloy. Several classification techniques are compared with our proposal: Linear Discriminant Analysis, k-Nearest Neighbor, Multilayer Perceptron, Support Vector Machines and Naive Bayes. The results obtained show the robustness and higher performance of the C-MANTEC algorithm in comparison to the other artificial intelligence models, corroborating the utility of the constructive neural networks paradigm in the modelling pitting corrosion problem.