A Hybrid System for Automatic Infant Cry Recognition II
Carlos Alberto Reyes-García (Instituto Nacional de Astrofísica Óptica y Electrónica, Mexico), Sandra E. Barajas (Instituto Nacional de Astrofísica Óptica y Electrónica, Mexico), Esteban Tlelo-Cuautle (Instituto Nacional de Astrofísica Óptica y Electrónica, Mexico) and Orion Fausto Reyes-Galaviz (Universidad Autónoma de Tlaxcala, Mexico)
Copyright © 2009.
OnDemand Chapter PDF Download
Download link provided immediately after order completion
Instant access upon order completion.
Automatic Infant Cry Recognition (AICR) process is basically a problem of pattern processing, very similar to the Automatic Speech Recognition (ASR) process (Huang, Acero, Hon, 2001). In AICR first we perform acoustical analysis, where the crying signal is analyzed to extract the more important acoustical features, like; LPC, MFCC, etc. (Cano, Escobedo and Coello, 1999). The obtained characteristics are represented by feature vectors, and each vector represents a pattern. These patterns are then classified in their corresponding pathology (Ekkel, 2002). In the reported case we are automatically classifying cries from normal, deaf and asphyxiating infants. We use a genetic algorithm to find several optimal parameters needed by the Fuzzy Relational Neural Network FRNN (Reyes, 1994), like; the number of linguistic properties, the type of membership function, the method to calculate the output and the learning rate. The whole model has been tested on several data sets for infant cry classification. The process, as well as some results, is described.
Within the evolutionary techniques, perhaps one of the most popular is the genetic algorithm (AG) (Goldberg, 1989). Its structure presents analogies with the biological theory of evolution, and is based on the principle of the survival of the fittest individual (Holland, 1975). Generally, a genetic algorithm has five basic components (Michalewicz, 1992). A representation of potential solutions to the problem, a form to create potential initial solutions, a fitness function that is in charge to evaluate solutions, genetic operators that alter the offspring’s composition, and values for parameters like the size of the population, crossover probability, mutation probability, number of generations and others. Here we present different features of the genetic algorithm used to find a combination of parameters for the FRNN.
Complete Chapter List
Search this Book:
Fernando Zacarías Flores, Dionicio Zacarías Flores, Rosalba Cuapa Canto, Luis Miguel Guzmán Muñoz
Key Terms in this Chapter
Genetic Algorithms: A family of computational models inspired by evolution. These algorithms encode a potential solution to a specific problem on a simple chromosome-like data structure and apply recombination operators to these structures so as to preserve critical information. Genetic algorithms are often viewed as function optimizers, although the range of problems to which genetic algorithms have been applied is quite broad
Signal Processing: The analysis, interpretation and manipulation of signals. Processing of such signals includes storage and reconstruction, separation of information from noise, compression, and feature extraction
Fitness Function: It is a function defined over the genetic representation and measures the quality of the represented solution. The fitness function is always problem dependent.
Evolutionary Computation: A subfield of computational intelligence that involves combinatorial optimization problems. It uses iterative progress, such as growth or development in a population, which is then selected in a guided random search to achieve the desired end. Such processes are often inspired by biological mechanisms of evolution
Hybrid Intelligent System: A software system which employs, in parallel, a combination of methods and techniques mainly from subfields of Soft Computing
Binary Chromosome: Is an encoding scheme representing one potential solution to a problem, during a searching process, by means of a string of bits
Soft Computing: A partnership of techniques which in combination are tolerant of imprecision, uncertainty, partial truth, and approximation, and whose role model is the human mind. Its principal constituents are Fuzzy Logic (FL), Neural Computing (NC), Evolutionary Computation (EC) Machine Learning (ML) and Probabilistic Reasoning (PR)