Enrique Alexandre

Feature Selection for Sound Classification in Hearing Aids Through Restricted Search Driven by Genetic Algorithms

Hearing loss may disqualify many people from leading a normal life, though the majority do not make use of hearing aids. This is because most hearing aids on the market cannot automatically adapt to the changing acoustical environment the user faces daily. This paper focuses on the development of an automatic sound classifier for digital hearing aids that aims to enhance listening comprehension when the user goes from one sound environment to another. Given the strong complexity constraints of these devices, reducing the number of signal-describing features which feed the automatic classifier is of great importance and becomes a challenging topic. Thus, the use of genetic algorithms with restricted search is explored for the mentioned feature selection. In an effort to evaluate its performance, the algorithm is compared with a standard unconstrained genetic algorithm and with sequential methods. The restricted search driven by the implemented genetic algorithm performs better than both the sequential methods and unconstrained genetic algorithms. It thus allows a subset of signal-describing features with lower cardinality to be selected. This may permit these selected features to be programmed on the digital signal processor that the hearing aid is based on, and to make efficient use of its limited computational facilities.

Hybridizing Extreme Learning Machines and Genetic Algorithms to select acoustic features in vehicle classification applications

Currently traffic noise has become an important factor that affects human health, and thus, an application able to classify vehicles on the basis of the sound they produce becomes important in the effort of fulfilling recommendations that aim at reducing traffic noise and improving intelligent transportation systems. This paper focuses on the problem of selecting those sound-describing features that make the vehicle classifier work properly. In particular, the goal of this paper is to evaluate the feasibility of a novel feature selection method based on a special class of Genetic Algorithm (with restricted search) hybridized with a Extreme Learning Machine. Because of its great generalization performance at a very fast learning speed, the Extreme Learning Machine plays the key role of providing the fitness of candidate solutions in each generation of the Genetic Algorithm. After a number of experiments comparing its performance to that of other fast learning algorithms, our approach has been found to be the most feasible for the application at hand. The proposed method helps the Extreme Learning Machine-based classifier to increase its performance from a mean probability of correct classification of 74.83% (with no feature selection) up to 93.74% (when using the optimum subset of selected features).

Significant wave height and energy flux range forecast with machine learning classifiers

Abstract In this paper, the performance of different ordinal and nominal multi-class classifiers is evaluated, in a problem of wave energy range prediction using meteorological variables from numerical models. This prediction could be used in problems of wave energy conversion in renewable and sustainable systems for energy supply. Specifically, the work is focused on ordinal classifiers, that have provided excellent performance in previous applications. The proposed techniques are novel with respect to alternative classification and regression techniques used up to date, the former not considering the order relation between classes in a multi-class problem and the latter needing, in general, more complex models. Another important novelty of the paper is to consider meteorological variables from numerical models as inputs of the classifiers, which has not been done before, to our knowledge, in this context. For this, a data matching is carried out between meteorological data, obtained from NCEP/NCAR Reanalysis Project in four points around the two buoys subjected to study (a buoy in the Gulf of Alaska and another one in the Southeast of United States), and the wave height or wave period collected by sensors in each buoy. Using this matching, the problem is tackled as an ordinal multi-class classification problem and the objective is to predict the range of height of the wave produced in each buoy and the range of energy flux generated. The classifiers to be compared and the model proposed are fully evaluated in both buoys. The results obtained are promising, showing an acceptable reconstruction by ordinal methods with respect to nominal ones in terms of wave height and energy flux.

On the Use of 2-D Coding Techniques for ECG Signals

This paper discusses the convenience of using two-dimensional (2-D) coding techniques for the compression of electrocardiogram (ECG) signals. These signals present a very clear periodicity that can be exploited by the use of a 2-D time/frequency transform to decorrelate it as much as possible. A brief theoretical approach is given to justify the use of this technique, and a comparison is made between a 2-D and a one-dimensional (1-D) uniform quantization scenarios. The influence of the error as well as the frame size on the estimation of the fundamental period is studied

Sound Classification in Hearing Aids by the Harmony Search Algorithm

This chapter focuses on the application of the harmony search algorithms to the problem of selecting more appropriate features for sound classification in digital hearing aids. Implementing sound classification algorithms embedded in hearing aids is a very challenging task. Hearing aids have to work at very low clock frequency in order to minimize power consumption, and thus maximize battery life. This necessitates the reduction of computational load while maintaining a low error probability. Since the feature extraction process is one of the most time-consuming tasks, selecting a reduced number of appropriate features is essential, thus requiring low computational cost without degrading the operation. The music-inspired harmony-search (HS) algorithm allows for effectively searching adequate solutions to this strongly constrained problem. By starting with an initial set of 74 different sound-describing features, a number of experiments were carried out to test the performance of the proposed method. Results of the harmony search algorithm are compared to those reached by other widely used methods.

Reducing the computational cost for sound classification in hearing aids by selecting features via genetic algorithms with restricted search

This paper centers on designing a feature-selection algorithm able to provide a ldquosmallrdquo number of adequate features that assist a sound classification system for hearing aids in reducing its computational load without degrading its performance. Because of the problem complexity, we have explored the use of genetic algorithms with restricted search for the mentioned feature selection. In an effort to evaluate its performance, the algorithm has been compared to a standard unconstrained genetic algorithm and with sequential methods. The restricted search driven by the proposed algorithm performs better than both the sequential methods and unconstrained genetic algorithms. The proposed algorithm selects a feature subset composed of only 21 features, much smaller than the 76 features of the complete, original set of available features. This low-cardinality subset of signal-describing features is the one implemented on the hearing aid, saving thus a great number of the scarce computational resources, and making possible to put into practice the concept at reasonable cost.

Two-layer automatic sound classification system for conversation enhancement in hearing aids

This paper focuses on the development of an automatic sound classifier for digital hearing aids that aims to enhance the listening comprehension when the user goes from a sound environment to another different one. The approach consists in dividing the classifying algorithm into two layers that make use of two-class algorithms that work more efficiently: the input signal discriminated by the first layer into either speech or non-speech is ulteriorly classified more specifically depending on whether the user is in a conversation (both in quiet or in the presence of background noise) or in a noisy ambient in the absent of speech. The system results in having four classes, labeled speech in quiet, speech in noise, stationary noisy environments (for instance, an aircraft cabin), and non-stationary noisy environments. The combination of classifiers that has been found to be more successful in terms of probability of correct classification consists of a system that makes use of Multilayer Perceptrons for those classification tasks in which speech is involved, and a Fisher Linear Discrimnant for distinguising stationary noisy environments from the non-stationary ones. The system performance has been found to be higher than that of other more classical approaches, and even superior than that of our preliminary work.

Enhancing the energy efficiency of wireless-communicated binaural hearing aids for speech separation driven by soft-computing algorithms

Abstract: Assisted by soft computing methods, the work we present in this paper focuses on the design of energy-efficient algorithms for binaural hearing aids that aim to separate speech from other sounds the hearing impaired person is not interested in. To do this, the right and left hearing aids need to wirelessly transmit to each other some parameters involved in the speech separation algorithm. The problem is that this transmission appreciably reduces the battery life, the most important constrain for designing advanced algorithms in hearing aids. Reducing the number of bits used to represent the parameters to be transmitted will bring down the power consumption, but at the expense of degrading the ability of the system to separate the speech from the other sound sources. Aiming at solving this problem, our approach, based on quantizing the parameters to be transmitted, basically consists in computing the adequate number of quantization bits by means of a combination of neural networks and genetic algorithms in the effort of finding a balance between low bit rate (and thus, low power consumption) and good separation of speech. The results show that even by using only 2bits/quantized-sample, the quality of the separation is as high as 70% of the limiting non-quantized quality separation factor, which has been found to be 85%.

Fast communication: Joint design of Gaussianized spectrum-based features and least-square linear classifier for automatic acoustic environment classification in hearing aids

In this paper we propose a method to generate a novel set of features in order to improve sound classification in digital hearing aids. The approach is based on the fact that those classification algorithms whose design consists in minimizing the mean squared error work better when the data to be classified exhibit a Gaussian distribution. The novel features we propose are thus based on sound spectral magnitudes that, prior to the feature calculation itself, are Gaussianized by a power law parametrized by a design parameter, @a. The explored method allows to jointly design the sound features and a least-square linear classifier, whose design parameters are also parametrized by @a. The experimental work suggests that there is a proper value of @a for which the so-designed classifier, fed with the novel features, exhibits a low error probability. Moreover, we have found that the method can be extended to nonlinear classifiers also trained by minimizing the mean squared error, such as, for instance, neural networks.

Influence of acoustic feedback on the learning strategies of neural network-based sound classifiers in digital hearing aids

Sound classifiers embedded in digital hearing aids are usually designed by using sound databases that do not include the distortions associated to the feedback that often occurs when these devices have to work at high gain and low gain margin to oscillation. The consequence is that the classifier learns inappropriate sound patterns. In this paper we explore the feasibility of using different sound databases (generated according to 18 configurations of real patients), and a variety of learning strategies for neural networks in the effort of reducing the probability of erroneous classification. The experimental work basically points out that the proposed methods assist the neural network-based classifier in reducing its error probability in more than 18%. This helps enhance the elderly users comfort: the hearing aid automatically selects, with higher success probability, the program that is best adapted to the changing acoustic environment the user is facing.