Benedito G. Aguiar Neto

Parametric cepstral analysis for pathological voice assessment

Traditional methods to diagnose laryngeal pathologies such as laryngoscopy are considered invasive and uncomfortable. Methods based on acoustic analisys of speech signals have been investigated in order to diminish the number of laryngoscopical exams. Digital signal processing techniques have been used to perform an acoustic analysis for vocal quality assessment due to the simplicity and the non-invasive nature of the measurement procedures. Their employment is of special interest, as they can provide an objective diagnosis of pathological voices, and may be used as complementary tool in laryngoscopy. The degree of reliability and effectiveness of discriminating process of pathological voices from normal ones depends on the characteristics and parameters of voice used to train the employed classifier. This paper aims at evaluating the performance of the Linear Prediction Coding (LPC)-based cepstral analysis to discriminate pathological voices of speakers affected by vocal fold edema. For this purpose, LPC, cepstral, weighted cepstral, delta cepstral weighted delta cepstral mel-cepstral coefficients and are used. A vector-quantizing-trained distance classifier is used in the discrimination process.

A self-organizing algorithm for vector quantizer design applied to signal processing.

Vector quantization plays an important role in many signal processing problems, such as speech/speaker recognition and signal compression. This paper presents an unsupervised algorithm for vector quantizer design. Although the proposed method is inspired in Kohonen learning, it does not incorporate the classical definition of topological neighborhood as an array of nodes. Simulations are carried out to compare the performance of the proposed algorithm, named SOA (self-organizing algorithm), to that of the traditional LBG (Linde-Buzo-Gray) algorithm. The authors present an evaluation concerning the codebook design for Gauss-Markov and Gaussian sources, since the theoretic optimal performance bounds for these sources, as described by Shannons Rate-Distortion Theory, are known. In speech and image compression, SOA codebooks lead to reconstructed (vector-quantized) signals with better quality as compared to the ones obtained by using LBG codebooks. Additionally, the influence of the initial codebook in the algorithm performance is investigated and the algorithm ability to learn representative patterns is evaluated. In a speaker identification system, it is shown that the the codebooks designed by SOA lead to higher identification rates when compared to the ones designed by LBG.

Combining entropy measurements and cepstral analysis for pathological voice assessment

This work aims at investigating the combination of classifiers based on three different entropy measurements (Shannon, Relative and Tsallis) and four cepstral coefficients (cepstral, delta-cepstral, weighted cepstral and weighted delta cepstral) to discriminate pathological voices under Reinkes Edema from normal ones. The performance of the combined classifiers is evaluated considering the average and product rules. Classification accuracy is improved when combinations of acoustic features were considered, compared to individual classifiers results.

LPC modelling and cepstral analysis applied to vocal fold pathology detection

Laryngeal pathologies are generally diagnosed using laryngoscopical exams, which are considered invasive to patients. Digital signal processing techniques are noninvasive and can be applied to perform an acoustic analysis for vocal quality assessment providing an objective diagnosis of pathological voices. This paper aims at specifying and evaluating the acoustic features for vocal fold edema through a parametric modelling based on the resonant structure of the human speech production mechanism by LPC and LPC-based cepstral coefficients and a nonparametric approach related to human auditory perception system by mel-frequency cepstral coefficients. A vector-quantising-trained distance classifier is used in the discrimination process.

Evaluation of the impact in reducing the number of parameters for continuous speech recognition for Brazilian Portuguese

With the advent of technology, machines predominate in almost all scenarios of everyday life. The possibility of performing human-machine communication through speech makes this interaction easier and more productive. However, processing requirements still make difficult the implementation of systems for automatic speech recognition on devices with low computational power such as mobile phones, palmtops and appliances. For this purpose, it becomes necessary to research for optimized acoustic and language modeling, associated with use of representative databases, looking for a good compromise between recognition rates and computational demands imposed by embedded systems. The main goal of this work is to model the architecture of a system for continuous speech recognition to Brazilian Portuguese, in order to enable its implementation in an embedded system with limited computing resources.

Pathological voice discrimination based on entropy measurements

Digital signal processing techniques have been used to analyze vocal disorders caused by laryngeal pathologies. Laryngoscopical exams commonly used for detection of these diseases are invasive methods that cause discomfort to the patients. Because of its noninvasive nature, acoustical analysis of the temporal and spectral features of the voice signal can be used as an auxiliary technique in laringoscopical exams. This analysis may be applied for detection of vocal diseases and the evaluation of the vocal quality of patients subjected to surgical processes or medical treatments in the vocal folds. This work aims at investigating the behavior of entropy measures in disordered voices influenced by the presence of pathologies in the vocal folds. For this purpose, the Shannon entropy and the relative entropy are implemented, and their behavior for normal and pathological voices affected by vocal fold edemas is observed. The measurement of the entropy efficiency in discriminating between normal and pathological voices is obtained.

Construção de Dicionários Voltados para a Redução da Complexidade Computacional da Etapa de Codificação da Quantização Vetorial

This work presents a methodology for designing codebooks for the purpose of reducing the computational complexity of the vector quantization (VQ) encoding phase. The methodology consists on efficiently exploring a symmetry observed in some signals. The proposed methodology is evaluated by simulations involving voice waveform coding based on VQ.

Aprendizagem Competitiva versus Algoritmo LBG Quanto à Complexidade Computacional

Codebook design plays a crucial role in the performance of signal processing systems based on vector quantization (VQ). This paper is devoted to the comparison of a competitive learning algorithm with the traditional LBG (Linde-Buzo-Gray) algorithm in terms of computational complexity. From analytical expressions for the number of operations performed by those algorithms for VQ codebook design, the authors establish the conditions that may be satisfied so that the competitive algorithm be more efficient than the LBG algorithm in terms of the number of operations (multiplications, divisions, additions, subtractions and comparisons) performed.

Quantizao Vetorial Robusta Utilizando Simulated Annealing

Thepresentpaper is concernedwith the application of simulatedannealing(SA) to robust vector quantizer design,that is, SAis usedto providea suitableindex assignmetto the reconstructionvectors of a vectorquantization codebook,asan attemptto reducethe effectsof channelerrors. Simulationresultsregardingimagetransmissionover a binary symmetricchannelshowthat an imagequality improvementis obtainedbyusingSAindexing.

A signal generator for real time vibrotactile stimulations

A compact and reduced cost signal generator with variable frequency is presented in this work. It is programmed through serial communication, with application in systems of real time training for hearing impaired people with vibrotactile feedback. This signal generator was developed in order to correct delays caused by digitalization and voice signal processing in real time applications. Using an external generator, one separates the action or signal generation, to be applied in the tactile or vibrotactile actuator, from the acquisition and voice signal processing part.