Artificial Intelligence-Based Voice Assessment of Patients with Parkinson’s Disease Off and On Treatment: Machine vs. Deep-Learning Comparison

Sensors, Journal Year: 2023, Volume and Issue: 23(4), P. 2293 - 2293

Published: Feb. 18, 2023

Parkinson’s Disease (PD) is one of the most common non-curable neurodegenerative diseases. Diagnosis achieved clinically on basis different symptoms with considerable delays from onset processes in central nervous system. In this study, we investigated early and full-blown PD patients based analysis their voice characteristics aid commonly employed machine learning (ML) techniques. A custom dataset was made hi-fi quality recordings vocal tasks gathered Italian healthy control subjects patients, divided into diagnosed, off-medication hand, mid-advanced treated L-Dopa other. Following current state-of-the-art, several ML pipelines were compared usingdifferent feature selection classification algorithms, deep also explored a CNN architecture. Results show how feature-based achieve comparable results terms classification, KNN, SVM naïve Bayes classifiers performing similarly, slight edge for KNN. Much more evident predominance CFS as best selector. The selected features act relevant biomarkers capable differentiating subjects, untreated patients.

Language: Английский

---

Machine learning- and statistical-based voice analysis of Parkinson’s disease patients: A survey

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 219, P. 119651 - 119651

Published: Feb. 2, 2023

Language: Английский

---

High-Level CNN and Machine Learning Methods for Speaker Recognition

Sensors, Journal Year: 2023, Volume and Issue: 23(7), P. 3461 - 3461

Published: March 25, 2023

Speaker Recognition (SR) is a common task in AI-based sound analysis, involving structurally different methodologies such as Deep Learning or "traditional" Machine (ML). In this paper, we compared and explored the two on DEMoS dataset consisting of 8869 audio files 58 speakers emotional states. A custom CNN to several pre-trained nets using image inputs spectrograms Cepstral-temporal (MFCC) graphs. AML approach based acoustic feature extraction, selection multi-class classification by means Naïve Bayes model also considered. Results show how custom, less deep trained grayscale spectrogram images obtain most accurate results, 90.15% 83.17% colored MFCC. AlexNet provides comparable reaching 89.28% 83.43% MFCC.The classifier 87.09% accuracy 0.985 average AUC while being faster train more interpretable. Feature shows F0, MFCC voicing-related features are characterizing for SR task. The high amount training samples content better reflect real case scenario speaker recognition, account generalization power models.

Language: Английский

---

Speech emotion classification using attention based network and regularized feature selection

Scientific Reports, Journal Year: 2023, Volume and Issue: 13(1)

Published: July 25, 2023

Speech emotion classification (SEC) has gained the utmost height and occupied a conspicuous position within research community in recent times. Its vital role Human-Computer Interaction (HCI) affective computing cannot be overemphasized. Many primitive algorithmic solutions deep neural network (DNN) models have been proposed for efficient recognition of from speech however, suitability these methods to accurately classify with multi-lingual background other factors that impede is still demanding critical consideration. This study an attention-based pre-trained convolutional regularized neighbourhood component analysis (RNCA) feature selection techniques improved emotion. The attention model proven successful many sequence-based time-series tasks. An extensive experiment was carried out using three major classifiers (SVM, MLP Random Forest) on publicly available TESS (Toronto English Sentence) dataset. result our (Attention-based DCNN+RNCA+RF) achieved 97.8% accuracy yielded 3.27% performance, which outperforms state-of-the-art SEC approaches. Our evaluation revealed consistency mechanism human behavioural patterns classifying auditory speech.

Language: Английский

---

MTLSER: Multi-task learning enhanced speech emotion recognition with pre-trained acoustic model

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: 273, P. 126855 - 126855

Published: Feb. 18, 2025

Language: Английский

---

Automatic Speech Emotion Recognition of Younger School Age Children

Mathematics, Journal Year: 2022, Volume and Issue: 10(14), P. 2373 - 2373

Published: July 6, 2022

This paper introduces the extended description of a database that contains emotional speech in Russian language younger school age (8–12-year-old) children and describes results validation based on classical machine learning algorithms, such as Support Vector Machine (SVM) Multi-Layer Perceptron (MLP). The is performed using standard procedures scenarios similar to other well-known databases children’s acting speech. Performance evaluation automatic multiclass recognition four emotion classes “Neutral (Calm)—Joy—Sadness—Anger” shows superiority SVM performance also MLP over perceptual tests. Moreover, test dataset which was used are even better. These prove emotions can be reliably recognized both by experts automatically algorithms MLP, baselines for comparing systems more sophisticated modern methods deep neural networks. confirm this valuable resource researchers studying affective reactions communication during child-computer interactions develop various edutainment, health care, etc. applications.

Language: Английский

---

Detecting multiple coexisting emotions from public emergency opinions

Journal of Information Science, Journal Year: 2024, Volume and Issue: unknown

Published: Feb. 21, 2024

To detect multiple coexisting emotions from public emergency opinions, this article proposes a novel two-stage emotion-detection model. First, the text semantic feature extracted through bidirectional encoder representation transformers (BERT) and emotion lexicon dictionary are fused. Then, subjectivity judgement detection performed in two separate stages. In first stage, we introduce synthetic minority oversampling technique (SMOTE) to enhance balance of data distribution select optimal classifier recognise opinion texts with emotion. second label powerset (LP)-SMOTE is proposed increase number category samples, multichannel classifiers decision mechanism employed different types determine final labels. Finally, Weibo about coronavirus disease 2019 (COVID-19) collected verify effectiveness Experiment results indicate that model outperforms state-of-the-art models, F1_macro 0.8532, F1_micro 0.8333, hamming loss 0.0476. The conducive decision-making for departments.

Language: Английский

---

Multimodal Affective Communication Analysis: Fusing Speech Emotion and Text Sentiment Using Machine Learning

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(15), P. 6631 - 6631

Published: July 29, 2024

Affective communication, encompassing verbal and non-verbal cues, is crucial for understanding human interactions. This study introduces a novel framework enhancing emotional by fusing speech emotion recognition (SER) sentiment analysis (SA). We leverage diverse features both classical deep learning models, including Gaussian naive Bayes (GNB), support vector machines (SVMs), random forests (RFs), multilayer perceptron (MLP), 1D convolutional neural network (1D-CNN), to accurately discern categorize emotions in speech. further extract text from speech-to-text conversion, analyzing it using pre-trained models like bidirectional encoder representations transformers (BERT), generative transformer 2 (GPT-2), logistic regression (LR). To improve individual model performance SER SA, we employ an extended dynamic Bayesian mixture (DBMM) ensemble classifier. Our most significant contribution the development of two-layered DBMM (2L-DBMM) multimodal fusion. effectively integrates sentiment, enabling classification more nuanced, second-level states. Evaluating our on EmoUERJ (Portuguese) ESD (English) datasets, achieves accuracy rates 96% 98% SER, 85% 95% combined 2L-DBMM, respectively. findings demonstrate superior modalities compared classifiers 2L-DBMM merging different modalities, highlighting value methods fusion affective communication analysis. The results underscore potential approach with broad applications fields mental health assessment, human–robot interaction, cross-cultural communication.

Language: Английский

---

Voice Disorder Multi-Class Classification for the Distinction of Parkinson’s Disease and Adductor Spasmodic Dysphonia

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(15), P. 8562 - 8562

Published: July 25, 2023

Parkinson’s Disease and Adductor-type Spasmodic Dysphonia are two neurological disorders that greatly decrease the quality of life millions patients worldwide. Despite this great diffusion, related diagnoses often performed empirically, while it could be relevant to count on objective measurable biomarkers, among which researchers have been considering features voice impairment can useful indicators but sometimes lead confusion. Therefore, here, our purpose was aimed at developing a robust Machine Learning approach for multi-class classification based 6373 extracted from convenient dataset made sustained vowel/e/ an ad hoc selected Italian sentence, by 111 healthy subjects, 51 disease patients, 60 dysphonic patients. Correlation, Information Gain, Gain Ratio, Genetic Algorithm-based methodologies were compared feature selection, build subsets analyzed means Naïve Bayes, Random Forest, Multi-Layer Perceptron classifiers, trained with 10-fold cross-validation. As result, spectral, cepstral, prosodic, voicing-related assessed as most relevant, Algorithm effective selector, adopted classifiers similarly. In particular, + Bayes brought one highest accuracies in analysis, being 95.70% vowel 99.46% sentence.

Language: Английский

---

Acoustic Analysis of Speech for Screening for Suicide Risk: Machine Learning Classifiers for Between- and Within-Person Evaluation of Suicidality

Journal of Medical Internet Research, Journal Year: 2023, Volume and Issue: 25, P. e45456 - e45456

Published: Feb. 26, 2023

Assessing a patient's suicide risk is challenging for health professionals because it depends on voluntary disclosure by the patient and often has limited resources. The application of novel machine learning approaches to determine clinical utility.This study aimed investigate cross-sectional longitudinal assess suicidality based acoustic voice features psychiatric patients using artificial intelligence.We collected 348 recordings during interviews 104 diagnosed with mood disorders at baseline 2, 4, 8, 12 months after recruitment. Suicidality was assessed Beck Scale Suicidal Ideation suicidal behavior Columbia Suicide Severity Rating Scale. voice, including temporal, formal, spectral features, were extracted from recordings. A between-person classification model that examines vocal characteristics individuals cross sectionally detect high within-person detects considerable worsening changes in within an individual developed compared. Internal validation performed 10-fold audio data 2-month external 2 4 months.A combined set 3 demographic variables (age, sex, past attempts) included single-layer neural network model. Furthermore, 13 extreme gradient boosting algorithm classifier able 69% accuracy (sensitivity 74%, specificity 62%, area under receiver operating characteristic curve 0.62), whereas predict over 79% 68%, 84%, 0.67). second showed 62% predicting increased sets.Within-person analysis promising approach suicidality. Automated can be used support real-time assessment primary care or telemedicine.

Language: Английский

---