Pairwise Feature Research Articles

Speech stress classification and feature analysis using a neural network classifier

Variability in speech production due to task induced stress contributes significantly to loss in speech processing performance. A new neural network algorithm is formulated which classifies speech under stress using the susas stress database. Five spectral parameter representations (Mel, Δ-Mel, Δ2-Mel, Auto-Corr-Mel, and Cross-Corr-Mel cepstral) are considered as potential stressed speech relayers. Eleven stress conditions are considered which include Angry, Clear, Fast, Lombard, Loud, Normal, Question, Slow, and Soft speech. Stressed speech classification using these features is evaluated with respect to (i) pairwise class separability, (ii) an objective measure of pairwise and global feature separability, and (iii) analysis of articulatory vocal tract area functions. Perturbations in speech production under stress are reflected to varying degrees in the five speech feature representations. A neural network classifier over phoneme partitions can achieve good speaker dependent classification performance (80.6%) when stress conditions are combined into six groups of related stress domains. The implication of reliable stress classification will be discussed for applications such as monitoring speaker state, improving naturalness of speech coding, and increasing robustness of speech recognizers in adverse conditions.

Conceptual Clustering, Categorization, and Polymorphy

In this paper we describe WITT, a computational model of categorization and conceptual clustering that has been motivated and guided by research on human categorization. Properties of categories to which humans are sensitive include best or prototypical members, relative contrasts between categories, and polymorphy (neither necessary nor sufficient feature rules). The system uses pairwise feature correlations to determine the “similarity” between objects and clusters of objects, allowing the system a flexible representation scheme that can model common-feature categories and polymorphous categories. This intercorrelation measure is cast in terms of an information-theoretic evaluation function that directs WITT'S search through the space of clusterings. This information-theoretic similarity metric also can be used to explain basic-level and typicality effects that occur in humans. WITT has been tested on both artificial domains and on data from the 1985 World Almanac, and we have examined the effect of various system parameters on the quality of the model's behavior.

Critical exponents for power-spectra scaling at mergings of chaotic bands.

The properties of the power spectra of a broad class of smooth d-dimensional (d\ensuremath{\ge}2) dynamical systems exhibiting a reverse sequence of pairwise mergings of chaotic bands is discussed. In particular, the width of the power pairwise spectral feature associated with the pairwise merging of chaotic bands scales as \ensuremath{\Vert}p-${p}_{m}$${\ensuremath{\Vert}}^{\ensuremath{\gamma}}$, where p is a system parameter, the mth merging occurs at p=${p}_{m}$, and \ensuremath{\gamma} is the critical exponent. Numerical experiments are performed which verify theoretical predictions for \ensuremath{\gamma}. For example, near the accumulation of period doublings the result for the critical exponent reduces to \ensuremath{\gamma}${=(1/2+2}^{\mathrm{\ensuremath{-}}m}$\ensuremath{\eta}/ln(1/J), where \ensuremath{\eta} is a universal number, J is the Jacobian determinant, and ${2}^{m}$ is the number of bands (before merging). (In the trivial case of smooth one-dimensional maps J\ensuremath{\rightarrow}0 and the exponent is (1/2.) .AE

Mortality of middle aged white South African gold miners.

A cohort of 3971 white miners in South Africa, born between 1 January 1916 and 31 December 1930 who were alive on 1 January 1970 and currently working in the East Rand-Central Rand-West Rand mining areas, was followed up for nine years, when the 3426 survivors were aged from 48 to 62. Fifteen (0.4%) had been lost to view and 530 had died (13.4% of the 3956 whose vital status was determined). Based on the occupational histories of a 30% sample of the cohort it was known that the vast majority were gold miners. An estimated 93% had worked more than 85% of their mining service in gold mines. Standardised mortality ratios were calculated as the ratios of the deaths observed in the cohort to those expected on the basis of concurrent mortality in the reference population--the total white male population in the Republic of South Africa. There was little sign of a "healthy worker effect"; of several possible reasons, one is that the white miner in South Africa had adopted certain unhealthy life styles, another is that the reference population was otherwise inappropriate. The SMR for all causes of death (117.6) was raised because of excess mortality due to the following causes: lung cancer (161.2), chronic respiratory diseases (165.6), and acute and chronic nephritis (381.0). A case-referent analysis was carried out on those miners in the cohort who had spent at least 85% of their service in gold mines. For lung cancer, smoking was the main contributory factor towards disease. For chronic respiratory diseases bronchitis, emphysema, asthma, pneumoconiosis, and pulmonary heart disease), smoking was also the main risk factor, but there was an association wih cumulative dust exposure. Raised blood pressure, smoking, and adiposity were associated with ischaemic heart disease as was the duration of service underground. Study of comprehensive medical histories in all 530 deaths, including necropsy in most cases, showed that none was directly due to pneumoconiosis or to tuberculosis.

A QSAR Study of the Ames Mutagenicity of 1‐(X‐Phenyl)‐3,3‐dialkyltriazenes Using Molecular Potential Energy Fields and Molecular Shape Analysis

AbstractA set of 18 1‐(X‐phenyl)‐3,3‐dialkyltriazenes whose mutagenic potencies have been measured were investigated in terms of their potential energy fields. Molecular shape analysis, MSA, was applied to the potential energy field data and a QSAR established. The MSA descriptor, common overlap steric volume between pairs of molecules, found to be a correlate in a previous study [1] is replaced in this study by what physically appears to be the difference in dispersion binding energy between pairs of molecules. A separate investigation of the interrelationship between molecular dielectric representation and corresponding potential energy field based MSA descriptors was carried out. Single molecule MSA descriptors derived from electrostatic energetics are quite dependent on dielectric form, while corresponding pairwise difference MSA features exhibit only a marginal relationship.