Speech emotion recognition (SER) has become a major area of investigation in human-computer interaction. Conventionally, SER is formulated as a classification problem that follows a common methodology: (i) extracting features from speech signals; and (ii) constructing an emotion classifier using extracted features. With the advent of deep learning, however, the former stage is integrated into the latter. That is to say, deep neural networks (DNNs), which are trained using log-Mel spectrograms (LMS) of audio waveforms, extract discriminative features from LMS. A critical issue, and one that is often overlooked, is that this procedure is done without relating the choice of LMS parameters to the performance of the trained DNN classifiers. It is commonplace in SER studies that practitioners assume some “usual” values for these parameters and devote major efforts to training and comparing various DNN architectures. In contrast with this common approach, in this work we choose a single lightweight pre-trained architecture, namely, SqueezeNet, and shift our main effort into tuning LMS parameters. Our empirical results using three publicly available SER datasets show that: (i) parameters of LMS can considerably affect the performance of DNNs; and (ii) by tuning LMS parameters, highly competitive classification performance can be achieved. In particular, treating LMS parameters as hyperparameters and tuning them led to ~ 23%, ~ 10%, and ~ 11% improvement in contrast with the use of “usual” values of LMS parameters in EmoDB, IEMOCAP, and SAVEE datasets, respectively.
20 October 2025 in The European Physical Journal C
We consider a modified gravity model with a running gravitational constant coupled to a varying dark energy fluid and test its imprint on the growth of structure in the universe. Using redshift space distortion (RSD) measurement results, we show a tension at the $$3 \sigma $$ level between the best fit $$\Lambda $$ CDM and the corresponding parameters obtained from the Planck data. Unlike many modified gravity-based solutions that overlook scale dependence and model-specific background evolution, we study this problem in the broadest possible context by incorporating both factors into our investigation. We performed a full perturbation analysis to demonstrate a scale dependence in the growth equation. Fixing the scale to $$k = 0.1 h$$ Mpc $$^{-1}$$ and introducing a phenomenological functional form for the varying Newton coupling G with only one free parameter, we conduct a likelihood analysis of the RSD selected data. The analysis reveals that the model can bring the tension level within $$1 \sigma $$ while maintaining the deviation of G from Newton’s gravitational constant at the fifth order.
Background: The rapid onset of COVID-19 placed immense strain on many already overstretched healthcare systems. The unique physiological changes in pregnancy, amplified by the complex effects of COVID-19 in pregnant women, rendered prioritization of infected expectant mothers more challenging. This work aims to use state-of-the-art machine learning techniques to predict whether a COVID-19-infected pregnant woman will be admitted to ICU (Intensive Care Unit). Methods: A retrospective study using data from COVID-19-infected women admitted to one hospital in Astana and one in Shymkent, Kazakhstan, from May to July 2021. The developed machine learning platform implements and compares the performance of eight binary classifiers, including Gaussian naïve Bayes, K-nearest neighbors, logistic regression with L2 regularization, random forest, AdaBoost, gradient boosting, eXtreme gradient boosting, and linear discriminant analysis. Results: Data from 1292 pregnant women with COVID-19 were analyzed. Of them, 10.4% were admitted to ICU. Logistic regression with L2 regularization achieved the highest F1-score during the model selection phase while achieving an AUC of 0.84 on the test set during the evaluation stage. Furthermore, the feature importance analysis conducted by calculating Shapley Additive Explanation values points to leucocyte counts, C-reactive protein, pregnancy week, and eGFR and hemoglobin as the most important features for predicting ICU admission. Conclusions: The predictive model obtained here may be an efficient support tool for prioritizing care of COVID-19-infected pregnant women in clinical practice.
Background: The rapid onset of COVID-19 placed immense strain on many already overstretched healthcare systems. The unique physiological changes of pregnancy, amplified by the complex effects of COVID-19 in pregnant women, rendered prioritization of infected expectant mothers more challenging. This work aims to use state-of-the-art machine learning techniques to predict whether a COVID-19-infected pregnant woman will be admitted to ICU (Intensive Care Unit). Methods: A retrospective study using data from COVID-19 infected women admitted to 2 hospital 1 in Astana and 1 in Shymkent, Kazakhstan, from May to July 2021. The developed machine learning platform implements and compares the performance of eight binary classifiers including Gaussian naïve Bayes, K-nearest neighbors, logistic regression with L2 regularization, random forest, AdaBoost, gradient boosting, eXtreme gradient boosting, and linear discriminant analysis. Results: Data from 1168 pregnant women with COVID-19 was analyzed. From them, 9.4% were admitted to ICU. Logistic regression with L2 regularization achieved the highest F1-score during the model selection phase while achieving an AUC of 0.84 on the test set during the evaluation stage. Furthermore, the feature importance analysis conducted by calculating Shapley Additive Explanation values points to leucocyte counts, C-reactive protein, pregnancy week, and eGFR and hemoglobin as the most important features for predicting ICU admission. Conclusion: The predictive model here obtained may be an efficient support tool for prioritizing care of COVID-19 infected pregnant women in clinical practice.