Ovidiu Grigore

Statistical and Hybrid Methods for Speech Recognition in Romanian

The present paper describes the evolution of our work concerning the problem of speech recognition. Beginning with a classical hidden Markov model (HMM), we have investigated two ways to improve the performance of this basic structure. The first way was to realize a neuro-statistical hybrid by integrating a multilayer perceptron (MLP) as a posteriori probability estimator. The system was further refined by adding supplementary discriminative training (DT) based on the minimum classification error (MCE). Tests performed on a 15,000 isolated spoken-word database, showed an increase in the recognition rate from 92.2% for the HMM-based recognition system, to 94.7% for the HMM-MLP system, and then to 98.1% for the refined HMM-MLP-DT system. The second way to improve the classical HMM was to build a fuzzy-statistical hybrid, FHMM, based on a fuzzy similarity measure instead of the probabilistic measure specific to the usual statistical model. The benefits of the fuzzy measure introduction were evaluated on a vowel recognition task, and a decrease of approximately 3% in the error rate is reported.

Pheromone averaging exploration algorithm

In this paper we introduce a novel spatial exploration and coverage algorithm based on reflex agents using a pheromone map as storage and communication medium. The algorithm proposed in this paper outperforms many of the popular reflex agent exploration algorithms in terms of exploration performance measured as the cumulative path length.

Syntactical Self-Organizing Map

In this paper a new neural network structure, called Syntactical Self-Organizing Map (SSOM), is introduced. SSOM is obtained from classical (numerical) Kohonen neural network and is specifically for classifying the syntactical structures, like: strings, trees or graphs. After defining the SSOM structure and algorithm, in the third part of the paper an application of character recognition is solved using SSOM. To point out the performances of the new neural network, a comparison of results obtained using the SSOM and the Fu and Lus clustering algorithm [10] for the same application is done. Moreover, we show that the syntactical Kohonen map have also the topological feature like the numerical one.

Intelligent System for Emotion Recognition Based on Electrodermal Activity Processing

This paper presents a simple method for detecting a psychological stressed or relaxed state by using only features extracted from the electrodermal activity signal. The signal is acquired with two small electrodes placed on the subject’s left hand. Current methodologies used for stress detection provide recognition rates between 40% and 71% while using multiple physiological signals. There are some published papers that describe stress detection methods with satisfactory results without providing a numerical recognition rate. A method used in previous studies to induce a relaxed or stressed state is also presented, which generates similar results in laboratory conditions for different subjects. Also, a very strict signal acquisition protocol is used with the purpose of minimizing artifacts caused by the bad electrode connection, environment or recording device.

Ant based exploration algorithms — A brief survey

In this paper we present a brief survey of the ant based multi-agent exploration algorithms and a performance comparison of these algorithms obtained by analyzing them in a variety of scenarios.

Impaired speech recognition Case study: Recognition of initial ‘r’ consonant in rhotacism affected pronunciations

The paper presents a study on one of the most common speech impairment in Romanian: the wrong pronunciation of the consonant “r”, called rhotacism. We propose a feature extraction method that can recognize rhotacism. The features extracted to characterize the interesting phonemes are the Mel-frequency cepstrum coefficients and their standard deviation over the signal duration. The final part of the paper presents a simple classification in correct or wrong pronounced of the resulting patterns based on the kNN algorithm and our feature extraction method. The system performance, expressed as accuracy in a classification experiment on a database containing correct and impaired pronunciations of the consonant “r”, is acceptable, validating the system to support the rhotacism diagnosis and therapy.

Lighting as Support for Enhancing Well-Being, Health and Mental Fitness of an Ageing Population - The FP6 EU Funded ALADIN Project

The paper presents the ALADIN prototype for adaptive lighting control designed to assist elderly in achieving a state of well-being, developed as a FP6 EU funded project. It uses psycho-physiological features extracted from Electro-Dermal Activity (EDA) and Pulse signals to determine the subjects mental state and adapts the lighting parameters in order to achieve a certain desired state. One of the controller implementations was done using Simulated Annealing. Field test evaluations of this implementation are discussed.