This paper presents a new Bayesian method for analyzing Ecological Momentary Assessment (EMA) data and applies this method in a re-analysis of data from a previous EMA study. The analysis method has been implemented as a freely available Python package EmaCalc, RRID:SCR 022943. The analysis model can use EMA input data including nominal categories in one or more situation dimensions, and ordinal ratings of several perceptual attributes. The analysis uses a variant of ordinal regression to estimate the statistical relation between these variables. The Bayesian method has no requirements related to the number of participants or the number of assessments by each participant. Instead, the method automatically includes measures of the statistical credibility of all analysis results, for the given amount of data. For the previously collected EMA data, the analysis results demonstrate how the new tool can handle heavily skewed, scarce, and clustered data that were collected on ordinal scales, and present results on interval scales. The new method revealed results for the population mean that were similar to those obtained in the previous analysis by an advanced regression model. The Bayesian approach automatically estimated the inter-individual variability in the population, based on the study sample, and could show some statistically credible intervention results also for an unseen random individual in the population. Such results may be interesting, for example, if the EMA methodology is used by a hearing-aid manufacturer in a study to predict the success of a new signal-processing method among future potential customers.

Objective: IOI-HA response data are conventionally analysed assuming that the ordinal responses have interval-scale properties. This study critically considers this assumption and compares the conventional approach with a method using Item Response Theory (IRT). Design: A Bayesian IRT analysis model was implemented and applied to several IOI-HA data sets. Study sample: Anonymised IOI-HA responses from 13273 adult users of one or two hearing aids in 11 data sets using the Australian English, Dutch, German and Swedish versions of the IOI-HA. Results: The raw ordinal responses to IOI-HA items do not represent values on interval scales. Using the conventional rating sum as an overall score introduces a scale error corresponding to about 10 − 15% of the true standard deviation in the population. Some interesting and statistically credible differences were demonstrated among the included data sets. Conclusions: It is questionable to apply conventional statistical measures like mean, variance, t-tests, etc., on the raw IOI-HA ratings. It is recommended to apply only nonparametric statistical test methods for comparisons of IOI-HA results between groups. The scale error can sometimes cause incorrect conclusions when individual results are compared. The IRT approach is recommended for analysis of individual results.

Mutual information (MI) is an important information theoretic concept which has many applications in telecommunications, in blind source separation, and in machine learning. More recently, it has been also employed for the instrumental assessment of speech intelligibility where traditionally correlation based measures are used. In this paper, we address the difference between MI and correlation from the viewpoint of discovering dependencies between variables in the context of speech signals. We perform our investigation by considering the linear predictive approximation and the extrapolation of speech signals as examples. We compare a parametric MI estimation approach based on a Gaussian mixture model (GMM) with the k-nearest neighbor (KNN) approach which is a well-known non-parametric method available to estimate the MI. We show that the GMM-based MI estimator leads to more consistent results.

This paper presents a method to analyze paired-comparison data including either binary or graded ordinal responses, with or without ties. The proposed method can use either of two classical choice models: (1) Thurstone case V, which assumes a Gaussian distribution of the sensory variables underlying listener decisions, or (2) the Bradley-Terry-Luce (BTL) model, which assumes a logistic distribution. The analysis method was validated using simulated paired-comparison experiments with known distributions of the sound-quality parameters in the simulated population from which "participants" were generated at random. The validation indicated that the Thurstone and BTL models give similar results close to the true values. The estimated credibility of a quality difference was slightly higher with the BTL model. The analysis results showed dramatically better precision when the response data included graded ordinal judgments instead of binary responses. Allowing tied responses also tended to improve precision. The method was also applied to data from a real evaluation of hearing-aid programs. The analysis revealed clinically interesting results with high statistical credibility, although the amount of test data was limited.

In this paper, we propose novel strategies for neutral vector variable decorrelation. Two fundamental invertible transformations, namely, serial nonlinear transformation and parallel nonlinear transformation, are proposed to carry out the decorrelation. For a neutral vector variable, which is not multivariate-Gaussian distributed, the conventional principal component analysis cannot yield mutually independent scalar variables. With the two proposed transformations, a highly negatively correlated neutral vector can be transformed to a set of mutually independent scalar variables with the same degrees of freedom. We also evaluate the decorrelation performances for the vectors generated from a single Dirichlet distribution and a mixture of Dirichlet distributions. The mutual independence is verified with the distance correlation measurement. The advantages of the proposed decorrelation strategies are intensively studied and demonstrated with synthesized data and practical application evaluations.

This paper presents a parametric Bayesian approach to the statistical analysis of phoneme confusion matrices measured for groups of individual listeners in one or more test conditions. Two different bias problems in conventional estimation of mutual information are analyzed and explained theoretically. Evaluations with synthetic datasets indicate that the proposed Bayesian method can give satisfactory estimates of mutual information and response probabilities, even for phoneme confusion tests using a very small number of test items for each phoneme category. The proposed method can reveal overall differences in performance between two test conditions with better power than conventional Wilcoxon significance tests or conventional confidence intervals. The method can also identify sets of confusion-matrix cells that are credibly different between two test conditions, with better power than a similar approximate frequentist method.

This paper addresses modelling data using the Watson distribution. The Watson distribution is one of the simplest distributions for analyzing axially symmetric data. This distribution has gained some attention in recent years due to its modeling capability. However, its Bayesian inference is fairly understudied due to difficulty in handling the normalization factor. Recent development of Markov chain Monte Carlo (MCMC) sampling methods can be applied for this purpose. However, these methods can be prohibitively slow for practical applications. A deterministic alternative is provided by variational methods that convert inference problems into optimization problems. In this paper, we present a variational inference for Watson mixture models. First, the variational framework is used to side-step the intractability arising from the coupling of latent states and parameters. Second, the variational free energy is further lower bounded in order to avoid intractable moment computation. The proposed approach provides a lower bound on the log marginal likelihood and retains distributional information over all parameters. Moreover, we show that it can regulate its own complexity by pruning unnecessary mixture components while avoiding over-fitting. We discuss potential applications of the modeling with Watson distributions in the problem of blind source separation, and clustering gene expression data sets.

Efficient quantization of the linear predictive coding (LPC) parameters plays a key role in parametric speech coding. The line spectral frequency (LSF) representation of the LPC parameters has found its applications in speech model quantization. In practical implementations of vector quantization (VQ), probability density function optimized VQ has been shown to be more efficient than the VQ based on training data. In this paper, we present the LSF parameters by a unit vector form, which has directional characteristics. The underlying distribution of this unit vector variable is modeled by a von Mises-Fisher mixture model (VMM). An optimal inter-component bit allocation strategy is proposed based on high rate theory and a distortion-rate (D-R) relation is derived for the VMM based-VQ (VVQ). Experimental results show that the VVQ outperforms the recently introduced Dirichlet mixture model-based VQ and the conventional Gaussian mixture model-based VQ in terms of modeling performance and D-R relation.

A novel Bayesian matrix factorization method for bounded support data is presented. Each entry in the observation matrix is assumed to be beta distributed. As the beta distribution has two parameters, two parameter matrices can be obtained, which matrices contain only nonnegative values. In order to provide low-rank matrix factorization, the nonnegative matrix factorization (NMF) technique is applied. Furthermore, each entry in the factorized matrices, i.e., the basis and excitation matrices, is assigned with gamma prior. Therefore, we name this method as beta-gamma NMF (BG-NMF). Due to the integral expression of the gamma function, estimation of the posterior distribution in the BG-NMF model can not be presented by an analytically tractable solution. With the variational inference framework and the relative convexity property of the log-inverse-beta function, we propose a new lower-bound to approximate the objective function. With this new lower-bound, we derive an analytically tractable solution to approximately calculate the posterior distributions. Each of the approximated posterior distributions is also gamma distributed, which retains the conjugacy of the Bayesian estimation. In addition, a sparse BG-NMF can be obtained by including a sparseness constraint to the gamma prior. Evaluations with synthetic data and real life data demonstrate the good performance of the proposed method.