Affective State during Physiotherapy and Its Analysis Using Machine Learning Methods.

Invasive or uncomfortable procedures especially during healthcare trigger emotions. Technological development of the equipment and systems for monitoring and recording psychophysiological functions enables continuous observation of changes to a situation responding to a situation. The presented study aimed to focus on the analysis of the individual's affective state. The results reflect the excitation expressed by the subjects' statements collected with psychological questionnaires. The research group consisted of 49 participants (22 women and 25 men). The measurement protocol included acquiring the electrodermal activity signal, cardiac signals, and accelerometric signals in three axes. Subjective measurements were acquired for affective state using the JAWS questionnaires, for cognitive skills the DST, and for verbal fluency the VFT. The physiological and psychological data were subjected to statistical analysis and then to a machine learning process using different features selection methods (JMI or PCA). The highest accuracy of the kNN classifier was achieved in combination with the JMI method (81.63%) concerning the division complying with the JAWS test results. The classification sensitivity and specificity were 85.71% and 71.43%.

Adolescent age estimation using voice features.

In this paper, a method for evaluating the chronological age of adolescents on the basis of their voice signal is presented. For every examined child, the vowels a, e, i, o and u were recorded in extended phonation. Sixty voice parameters were extracted from each recording. Voice recordings were supplemented with height measurement in order to check if it could improve the accuracy of the proposed solution. Predictor selection was performed using the LASSO (least absolute shrinkage and selection operator) algorithm. For age estimation, the random forest (RF) for regression method was employed and it was tested using a 10-fold cross-validation. The lowest absolute error (0.37 year ± 0.28) was obtained for boys only when all selected features were included into prediction. In all cases, the achieved accuracy was higher for boys than for girls, which results from the fact that the change of voice with age is larger for men than for women. The achieved results suggest that the presented approach can be employed for accurate age estimation during rapid development in children.

Prediction of menarcheal status of girls using voice features.

A method for evaluating the menarcheal status of girls on the basis of their voice features is presented in the paper. The registration procedure consists of voice recording and measuring 20 anthropological features. The input feature vector is a combination of voice and anthropometric parameters, counting 220 features. The optimal set of parameters was selected using five different methods: Method A - stepwise regression (first forward, then backward regression) performed on features with statistically different means/medians; Method B - stepwise regression (forward and backward) on all features, with age; Method C - stepwise regression as in B; including age, Method D - all features with statistically different means/medians, Method E - all features excluding age. For classification purposes three methods were employed: random forest (RF), support vector machine (SVM) and linear discriminant analysis (LDA) classifier. They were tested with 10-fold cross validation. The classification accuracy for RF using only voice features is higher than using only anthropometric data: 86.86% vs. 81.02% respectively. For the other two classifiers, the results do not show as large a difference: 80.60% vs. 82.80% for SVM and 80.66% vs. 82.34% for LDA. The advantage of voice features is more noticeable with sensitivity: 91.92% vs. 83.06% for RF. The obtained results suggest that the presented method can be used for automatic recognition of girls' menarcheal status using voice signal.

Dynamic time warping in phoneme modeling for fast pronunciation error detection.

The presented paper describes a novel approach to the detection of pronunciation errors. It makes use of the modeling of well-pronounced and mispronounced phonemes by means of the Dynamic Time Warping (DTW) algorithm. Four approaches that make use of the DTW phoneme modeling were developed to detect pronunciation errors: Variations of the Word Structure (VoWS), Normalized Phoneme Distances Thresholding (NPDT), Furthest Segment Search (FSS) and Normalized Furthest Segment Search (NFSS). The performance evaluation of each module was carried out using a speech database of correctly and incorrectly pronounced words in the Polish language, with up to 10 patterns of every trained word from a set of 12 words having different phonetic structures. The performance of DTW modeling was compared to Hidden Markov Models (HMM) that were used for the same four approaches (VoWS, NPDT, FSS, NFSS). The average error rate (AER) was the lowest for DTW with NPDT (AER=0.287) and scored better than HMM with FSS (AER=0.473), which was the best result for HMM. The DTW modeling was faster than HMM for all four approaches. This technique can be used for computer-assisted pronunciation training systems that can work with a relatively small training speech corpus (less than 20 patterns per word) to support speech therapy at home.