Voice Samples Research Articles

Can Voice Parameters Provide Cutoff Values to Predict Dysphagia in Individuals with Multiple Sclerosis?

In dysphagia assessment, along with well-defined measurements and signs, voice parameters can potentially support clinical decision as a marker, but more evidence is needed. This study aims to determine the voice parameters that can predict the risk of dysphagia and to determine optimal cutoff values in individuals with multiple sclerosis (IwMS). Seventy-six adults participated in the study, including 39 IwMS and 37 healthy individuals (HI). The study used the Dysphagia in Multiple Sclerosis Questionnaire (DYMUS), Gugging Swallowing Screen (GUSS), and Voice Handicap Index (VHI-10) and recorded voice samples using Praat programme. Voice recordings were taken pre- and post-swallowing. The voice parameters analysed are fundamental frequency (F0), standard deviation F0 (SD F0), jitter (local), shimmer (local), and harmonic-to-noise ratio (HNR). Roc analysis was performed to examine the diagnostic accuracy performance of the risk for dysphagia/penetration. The parameters of IwMS pre-swallowing differed significantly from those of HI on the VHI-10, DYMUS, GUSS scores, and jitter (local), shimmer (local), and HNR. IwMS but not HI exhibited significant differences in shimmer (local) and HNR between the pre- and post-swallowing measurements. In IwMS, GUSS revealed significant differences in shimmer (local) pre- and post-swallowing between the groups with and without dysphagia/penetration. In the ROC analysis results, the area under the curve (AUC) for shimmer (local) pre-swallowing was 73.1% (cutoff = 1.69); post-swallowing, it was 78.6% (cutoff = 1.57). In conclusion, IwMS can be associated with differences in shimmer (local) and HNR parameters, low quality of life-related to voice, and dysphagia/penetration risk. The AUC values for shimmer (local) in IwMS pre- and post-swallowing may help to strengthen diagnostic decisions of dysphagia risk.

Text and Sound-Based Feature Extraction and Speech Emotion Classification for Korean

Embracing the complexities of human emotions conveyed through speech, this study ventures into Speech Emotion Recognition (SER) within the human-computer interaction domain, leveraging cutting-edge artificial intelligence technologies. Focusing on the auditory attributes of speech, such as tone, pitch, and rhythm, the research introduces an innovative approach that amalgamates deep learning techniques with the A Learnable Frontend for Audio Classification (LEAF) algorithm and wav2vec 2.0 pre-trained on a large corpus, specifically targeting Korean voice samples. This methodology underlines the capacity of these technologies to process and decipher complex vocal expressions, aiming to elevate emotion classification precision notably. The exploration extends the horizons of SER by accentuating auditory emotion cues and aspires to enrich machine interactions to be more intuitive and empathetic across various applications like healthcare and customer service. The outcomes underscore the efficacy of transformer-based models, particularly wav2vec 2.0 and LEAF, in capturing the subtle emotional states expressed in speech, thereby affirming the importance of auditory cues over conventional visual and textual indicators. The study's implications for further research herald a promising trajectory for evolving AI systems adept at nuanced emotion detection, thereby forging pathways toward more natural and human-centric interactions between individuals and machines. This advancement is crucial for developing empathetic AI that can seamlessly integrate into our daily lives, understanding and reacting to human emotions in a way that mirrors human understanding and compassion.

A Machine Learning Approach for Early Identification of Prodromal Parkinson's Disease.

Parkinson's disease (PD) affects approximately 6 million people worldwide. Data analysis of early PD symptoms using machine learning (ML) models may provide an inexpensive, non-invasive, and simple method for the remote diagnosis of early PD. The aim of this project was to analyze voice, computer keystrokes, spiral drawings, and gait data involving PD patients and controls available in public databases using ML models and identify early PD characteristics that are more pronounced than others. An ML model was developed using Random Forest to analyze existing clinical data for PD patients, prodromal PD patients with REM (rapid eye movement) sleep behavior disorder (RBD) symptoms, and non-PD healthy controls. We reviewed and collected data from the UCI (University of California Irvine) Machine Learning Repository, PPMI (Parkinson's Progression Markers Initiative), and Kaggle databases. ML analysis was carried out on voice samples in PD and RBD patients, computer keystroke data, spiral drawings, and gait datasets. The ML prediction model developed may be helpful in improving risk prediction for PD, enabling early intervention and resource prioritization. The ML study suggests that voice analysis is the most robust test, followed by computer keystroke data, spiral drawings, and gait analysis, in that order. Voice is affected even in RBD patients, revealing that it is a sensitive and early measure of prodromal PD. The low accuracy of the analysis indicates that several PD-positive samples may remain undetected and unclassified. Combining all four features, that is, voice analysis, computer keystroke data, spiral drawings, and gait analysis, may improve the overall accuracy.

Deepfake Audio Detection System

Abstract: To ensure the authenticity of audio material, it is important to establish reliable detection tools that can trace the spread and use of deepfake technologies. This study focuses on the methodology of deepfake audio identification via Mel Frequency Cepstral Coefficients (MFCC) as features with Random Forest as a classifier. By extracting MFCC features from audio clips and using them in a Random Forest model, one can learn unique spectral properties that can help distinguish deepfakes from authentic audio. The Random Forest algorithm, famous for its quality of being able to work well in an ensemble learning paradigm, is utilized to identify patterns that are representative of deepfake manipulation. To ensure the efficiency and reliability of this method, it was tested on a large number of different data sets that included both genuine and fake voice samples. To ensure robustness and generalization, cross-validation techniques are employed, restricting model predictions to the range of 0 to 1 and providing informative error messages for effective diagnosis. Thus, it is an important scientific study helping to develop and strengthen methods for identifying and eliminating threats in the area of artificial sound activity.

Validity of Acoustic Measures Obtained Using Various Recording Methods Including Smartphones With and Without Headset Microphones.

The goal of this study was to assess various recording methods, including combinations of high- versus low-cost microphones, recording interfaces, and smartphones in terms of their ability to produce commonly used time- and spectral-based voice measurements. Twenty-four vowel samples representing a diversity of voice quality deviations and severities from a wide age range of male and female speakers were played via a head-and-thorax model and recorded using a high-cost, research standard GRAS 40AF (GRAS Sound & Vibration) microphone and amplification system. Additional recordings were made using various combinations of headset microphones (AKG C555 L [AKG Acoustics GmbH], Shure SM35-XLR [Shure Incorporated], AVID AE-36 [AVID Products, Inc.]) and audio interfaces (Focusrite Scarlett 2i2 [Focusrite Audio Engineering Ltd.] and PC, Focusrite and smartphone, smartphone via a TRRS adapter), as well as smartphones direct (Apple iPhone 13 Pro, Google Pixel 6) using their built-in microphones. The effect of background noise from four different room conditions was also evaluated. Vowel samples were analyzed for measures of fundamental frequency, perturbation, cepstral peak prominence, and spectral tilt (low vs. high spectral ratio). Results show that a wide variety of recording methods, including smartphones with and without a low-cost headset microphone, can effectively track the wide range of acoustic characteristics in a diverse set of typical and disordered voice samples. Although significant differences in acoustic measures of voice may be observed, the presence of extremely strong correlations (rs > .90) with the recording standard implies a strong linear relationship between the results of different methods that may be used to predict and adjust any observed differences in measurement results. Because handheld smartphone distance and positioning may be highly variable when used in actual clinical recording situations, smartphone + a low-cost headset microphone is recommended as an affordable recording method that controls mouth-to-microphone distance and positioning and allows both hands to be available for manipulation of the smartphone device.

A Methodology for Speaker Diazaration System Based on LSTM and MFCC Coefficients

Research on Speaker Identification is always difficult. A speaker may be automatically identified using by comparing their voice sample with their previously recorded voice, the machine learning strategy has grown in favor in recent years. Convolutional neural networks (CNN) , deep neural networks (DNN) are some of the machine learning techniques that has employed recently. The article will discuss a successful speaker verification system based on the d-vector to construct a new approach based on speaker diarization. In particular, in this article, we use the concept of LSTM to cluster the speech segments using MFCC coefficients and identify the speakers in the diarization system. The proposed system will be evaluated using benchmark performance metrics, and a comparative study will be made with other models. The need to consider the LSTM neural network using acoustic data and linguistic dialect is considered. LSTM networks could produce reliable speaker segmentation outputs.

Identifying bias in models that detect vocal fold paralysis from audio recordings using explainable machine learning and clinician ratings.

Detecting voice disorders from voice recordings could allow for frequent, remote, and low-cost screening before costly clinical visits and a more invasive laryngoscopy examination. Our goals were to detect unilateral vocal fold paralysis (UVFP) from voice recordings using machine learning, to identify which acoustic variables were important for prediction to increase trust, and to determine model performance relative to clinician performance. Patients with confirmed UVFP through endoscopic examination (N = 77) and controls with normal voices matched for age and sex (N = 77) were included. Voice samples were elicited by reading the Rainbow Passage and sustaining phonation of the vowel "a". Four machine learning models of differing complexity were used. SHapley Additive exPlanations (SHAP) was used to identify important features. The highest median bootstrapped ROC AUC score was 0.87 and beat clinician's performance (range: 0.74-0.81) based on the recordings. Recording durations were different between UVFP recordings and controls due to how that data was originally processed when storing, which we can show can classify both groups. And counterintuitively, many UVFP recordings had higher intensity than controls, when UVFP patients tend to have weaker voices, revealing a dataset-specific bias which we mitigate in an additional analysis. We demonstrate that recording biases in audio duration and intensity created dataset-specific differences between patients and controls, which models used to improve classification. Furthermore, clinician's ratings provide further evidence that patients were over-projecting their voices and being recorded at a higher amplitude signal than controls. Interestingly, after matching audio duration and removing variables associated with intensity in order to mitigate the biases, the models were able to achieve a similar high performance. We provide a set of recommendations to avoid bias when building and evaluating machine learning models for screening in laryngology.

Towards Early Intervention: Detecting Parkinson's Disease through Voice Analysis with Machine Learning

Introduction/ Background This study aims to utilize machine learning algorithms for early detection of Parkinson's Disease (PD) through voice recording analysis. Employing advanced machine learning techniques and a comprehensive dataset of voice samples, the objective is to develop a non-invasive, accurate, and reliable method for PD diagnosis, contributing to early intervention and management of the disease. Parkinson's Disease (PD) is a prevalent neurodegenerative disorder impacting millions globally. Early and accurate diagnosis is crucial for effective management and treatment. This study leverages Machine Learning (ML) algorithms to analyze voice recordings, aiming to improve PD detection. Materials and Methods We utilized a dataset of 195 voice samples with 23 attributes, applying machine learning algorithms such as Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Convolutional Neural Networks (CNN). The dataset was preprocessed, balanced, and evaluated using various performance metrics. Results The K-Nearest Neighbors (KNN) model demonstrated superior performance, achieving high precision (0.96-1.00), recall (0.97-1.00), and F1-scores (0.98-0.99) for both PD and non-PD classes, demonstrating an overall accuracy of 0.98 across 59 samples. This showcases its effectiveness in PD detection via voice analysis. Discussion This research underscores the potential of ML in revolutionizing PD detection through non-invasive methods. By comparing various algorithms, the study not only identifies the most effective model but also contributes to the broader understanding of applying ML techniques in healthcare. Conclusion The study's findings advocate for the KNN model as a promising tool for early and accurate PD diagnosis through voice analysis. The success of this model opens avenues for future research, including the exploration of more advanced algorithms and the integration of these models into practical diagnostic applications.

Deep-Learning-Based Segmentation of Predefined Chunks in Connected Speech: A Retrospective Analysis

Background and Objectives In institutional settings, manually segmenting connected speech is a time-consuming and labor-intensive process. This study aims to develop a deep-learning model for automating this process, evaluating its accuracy, and determining the minimum dataset size for effective performance.Materials and Method Voice data from 524 individuals with pathological conditions and 502 individuals with normal conditions, totaling 1026 samples, were used. Each voice sample had 17 chunks, including a “summer” sentence (15 chunks) and vowels /α/ and /i/. The deep-learning model employed in this study is based on the multi-layer perceptron-mixer architecture. This study evaluated performance using the Intersection over Union (IoU) metric, commonly employed in artificial intelligence-based image detection for chunk segmentation.Results The accuracy of chunk identification at the frame level was 96.47%. Using IoU metrics, chunk segmentation accuracy was 98.15% at IoU ≥0.6, 96.03% at IoU ≥0.7, and 89.78% at IoU ≥0.8. Optimal dataset size exploration indicated that more than 700 connected speech datasets were needed for successful training, maintaining F1-scores up to 95% at IoU ≥0.7.Conclusion The artificial intelligence model is suitable for the development of an automated system that efficiently divides segments in the institutional collection of voice data. This suggests its potential utility in advancing voice research using connected speech.

Classification of laryngeal diseases including laryngeal cancer, benign mucosal disease, and vocal cord paralysis by artificial intelligence using voice analysis

Voice change is often the first sign of laryngeal cancer, leading to diagnosis through hospital laryngoscopy. Screening for laryngeal cancer solely based on voice could enhance early detection. However, identifying voice indicators specific to laryngeal cancer is challenging, especially when differentiating it from other laryngeal ailments. This study presents an artificial intelligence model designed to distinguish between healthy voices, laryngeal cancer voices, and those of the other laryngeal conditions. We gathered voice samples of individuals with laryngeal cancer, vocal cord paralysis, benign mucosal diseases, and healthy participants. Comprehensive testing was conducted to determine the best mel-frequency cepstral coefficient conversion and machine learning techniques, with results analyzed in-depth. In our tests, laryngeal diseases distinguishing from healthy voices achieved an accuracy of 0.85–0.97. However, when multiclass classification, accuracy ranged from 0.75 to 0.83. These findings highlight the challenges of artificial intelligence-driven voice-based diagnosis due to overlaps with benign conditions but also underscore its potential.

Impact of Audio Data Compression on Feature Extraction for Vocal Biomarker Detection: Validation Study.

Vocal biomarkers, derived from acoustic analysis of vocal characteristics, offer noninvasive avenues for medical screening, diagnostics, and monitoring. Previous research demonstrated the feasibility of predicting type 2 diabetes mellitus through acoustic analysis of smartphone-recorded speech. Building upon this work, this study explores the impact of audio data compression on acoustic vocal biomarker development, which is critical for broader applicability in health care. The objective of this research is to analyze how common audio compression algorithms (MP3, M4A, and WMA) applied by 3 different conversion tools at 2 bitrates affect features crucial for vocal biomarker detection. The impact of audio data compression on acoustic vocal biomarker development was investigated using uncompressed voice samples converted into MP3, M4A, and WMA formats at 2 bitrates (320 and 128 kbps) with MediaHuman (MH) Audio Converter, WonderShare (WS) UniConverter, and Fast Forward Moving Picture Experts Group (FFmpeg). The data set comprised recordings from 505 participants, totaling 17,298 audio files, collected using a smartphone. Participants recorded a fixed English sentence up to 6 times daily for up to 14 days. Feature extraction, including pitch, jitter, intensity, and Mel-frequency cepstral coefficients (MFCCs), was conducted using Python and Parselmouth. The Wilcoxon signed rank test and the Bonferroni correction for multiple comparisons were used for statistical analysis. In this study, 36,970 audio files were initially recorded from 505 participants, with 17,298 recordings meeting the fixed sentence criteria after screening. Differences between the audio conversion software, MH, WS, and FFmpeg, were notable, impacting compression outcomes such as constant or variable bitrates. Analysis encompassed diverse data compression formats and a wide array of voice features and MFCCs. Wilcoxon signed rank tests yielded P values, with those below the Bonferroni-corrected significance level indicating significant alterations due to compression. The results indicated feature-specific impacts of compression across formats and bitrates. MH-converted files exhibited greater resilience compared to WS-converted files. Bitrate also influenced feature stability, with 38 cases affected uniquely by a single bitrate. Notably, voice features showed greater stability than MFCCs across conversion methods. Compression effects were found to be feature specific, with MH and FFmpeg showing greater resilience. Some features were consistently affected, emphasizing the importance of understanding feature resilience for diagnostic applications. Considering the implementation of vocal biomarkers in health care, finding features that remain consistent through compression for data storage or transmission purposes is valuable. Focused on specific features and formats, future research could broaden the scope to include diverse features, real-time compression algorithms, and various recording methods. This study enhances our understanding of audio compression's influence on voice features and MFCCs, providing insights for developing applications across fields. The research underscores the significance of feature stability in working with compressed audio data, laying a foundation for informed voice data use in evolving technological landscapes.

A Fine-Tuned CatBoost-Based Speech Disorder Detection Model

The classification of speech disorders (SDs) is crucial for treating children with speech impairment (SI). An automated SD classification can assist speech therapists in rendering services to children with SI in rural areas. Automated techniques for detecting SDs provide objective assessments of speech attributes, including articulation, fluency, and prosody. Clinical examinations and quantitative assessments provide an in-depth understanding of the patient’s speaking abilities and limitations. Existing deep learning (DL) models for SD detection often lack generalization across diverse populations and speech variations, leading to suboptimal performance when applied to individuals with different linguistic backgrounds or dialects. This study introduces a DL-based model for classifying normal and abnormal speeches using voice samples. To overcome the overfitting and bias, the authors construct convolutional neural network models with the weights of MobileNet V3 and EfficientNet B7 models for feature extraction (FE). To improve performance, they integrate the squeeze and excitation block with the MobileNet V3-based FE model. Similarly, the EfficientNet B7-model-based FE is improved using the structure pruning technique. The enhanced CatBoost model differentiates the normal and abnormal speeches using the extracted features. The experimental analysis is performed using the public dataset that contains 4620 utterances of healthy children and 2178 utterances of children with SI. The comparative study reveals the exceptional performance of the proposed SD classification model. The model outperforms the current SD classification models. It can be employed in clinical settings to support speech therapists. Substantial training with diverse voice samples can improve the generalizability of the proposed model.

Effects of Semi-occluded Vocal Tract Exercises with Different Vibration Sources and Duration on Healthy Adults’ Voice

Objectives: This study aimed to explore the impact of semi-occluded vocal tract exercises (SOVTE) with varied vibration sources and durations on the vocal performance of healthy adults. In addition, we sought to confirm the safety of voiced high-frequency oscillation (VHFO) intervention and provide basic data for performances in relation to straw phonation (SP).Methods: The study included 34 healthy adults with no vocal impairments, comprising 8 males and 26 females. The mean age of the participants was 23.00±2.06 years. To assess vocal characteristics during VHFO and SP, voice samples were recorded at three specific time points: prior to performing SOVTE (M0), after 1 minute of exercise (M1), and after 3 minutes of exercise (M3). Cepstral, perturbation, electroglottographic analyses were conducted on the recorded samples of vowel and sentences. Psychometric evaluation was conducted using questionnaires to gauge the self-reported severity of voice and vocal fatigue.Results: No statistically significant differences were observed in the average number of performances or duration per phonation between the two methods. When compared to baseline, CPPV significantly increased after 1 minute and 3 minutes of performance and Acoustic Psychometric Severity Index of Dysphonia decreased after 3 minutes of performance in VHFO. Both methods exhibited no significant differences in terms of psychometric and electroglottographic measurements.Conclusion: Results objectively confirmed the safety of VHFO intervention. This study is clinically significant as it lays the groundwork for future development of optimal treatment regimens and therapeutic programs for voice disorder patients, using foundational data on VHFO.

Longitudinal observational cohort study: Speech for Intelligent cognition change tracking and DEtection of Alzheimer’s Disease (SIDE-AD)

IntroductionThere is emerging evidence that speech may be a potential indicator and manifestation of early Alzheimer’s disease (AD) pathology. Therefore, the University of Edinburgh and Sony Research have partnered to...

StyleSinger: Style Transfer for Out-of-Domain Singing Voice Synthesis

Style transfer for out-of-domain (OOD) singing voice synthesis (SVS) focuses on generating high-quality singing voices with unseen styles (such as timbre, emotion, pronunciation, and articulation skills) derived from reference singing voice samples. However, the endeavor to model the intricate nuances of singing voice styles is an arduous task, as singing voices possess a remarkable degree of expressiveness. Moreover, existing SVS methods encounter a decline in the quality of synthesized singing voices in OOD scenarios, as they rest upon the assumption that the target vocal attributes are discernible during the training phase. To overcome these challenges, we propose StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference singing voice samples. StyleSinger incorporates two critical approaches for enhanced effectiveness: 1) the Residual Style Adaptor (RSA) which employs a residual quantization module to capture diverse style characteristics in singing voices, and 2) the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style attributes within the content representation during the training phase and thus improve the model generalization. Our extensive evaluations in zero-shot style transfer undeniably establish that StyleSinger outperforms baseline models in both audio quality and similarity to the reference singing voice samples. Access to singing voice samples can be found at https://stylesinger.github.io/.

Investigation of Deepfake Voice Detection Using Speech Pause Patterns: Algorithm Development and Validation

Background The digital era has witnessed an escalating dependence on digital platforms for news and information, coupled with the advent of “deepfake” technology. Deepfakes, leveraging deep learning models on extensive data sets of voice recordings and images, pose substantial threats to media authenticity, potentially leading to unethical misuse such as impersonation and the dissemination of false information. Objective To counteract this challenge, this study aims to introduce the concept of innate biological processes to discern between authentic human voices and cloned voices. We propose that the presence or absence of certain perceptual features, such as pauses in speech, can effectively distinguish between cloned and authentic audio. Methods A total of 49 adult participants representing diverse ethnic backgrounds and accents were recruited. Each participant contributed voice samples for the training of up to 3 distinct voice cloning text-to-speech models and 3 control paragraphs. Subsequently, the cloning models generated synthetic versions of the control paragraphs, resulting in a data set consisting of up to 9 cloned audio samples and 3 control samples per participant. We analyzed the speech pauses caused by biological actions such as respiration, swallowing, and cognitive processes. Five audio features corresponding to speech pause profiles were calculated. Differences between authentic and cloned audio for these features were assessed, and 5 classical machine learning algorithms were implemented using these features to create a prediction model. The generalization capability of the optimal model was evaluated through testing on unseen data, incorporating a model-naive generator, a model-naive paragraph, and model-naive participants. Results Cloned audio exhibited significantly increased time between pauses (P<.001), decreased variation in speech segment length (P=.003), increased overall proportion of time speaking (P=.04), and decreased rates of micro- and macropauses in speech (both P=.01). Five machine learning models were implemented using these features, with the AdaBoost model demonstrating the highest performance, achieving a 5-fold cross-validation balanced accuracy of 0.81 (SD 0.05). Other models included support vector machine (balanced accuracy 0.79, SD 0.03), random forest (balanced accuracy 0.78, SD 0.04), logistic regression, and decision tree (balanced accuracies 0.76, SD 0.10 and 0.72, SD 0.06). When evaluating the optimal AdaBoost model, it achieved an overall test accuracy of 0.79 when predicting unseen data. Conclusions The incorporation of perceptual, biological features into machine learning models demonstrates promising results in distinguishing between authentic human voices and cloned audio.

Symmetric and asymmetric Gaussian weighted linear prediction for voice inverse filtering

Weighted linear prediction (WLP) has demonstrated its significance in voice inverse filtering, contributing to enhanced methods for estimating both the vocal tract filter and the glottal source. WLP provides a mechanism to mitigate the effect on the linear prediction model of voice samples that affects the vocal tract filter estimation, particularly those samples around glottal closure instants (GCIs). This article studies the Gaussian weighted linear prediction (GLP) strategy, which employs a Gaussian attenuation window centered at the GCIs to reduce its contribution in the WLP analysis. In this study, the Gaussian attenuation is revisited and a parameterization of the window that adjusts to the typical variability in voice periodicity is introduced. In addition, an asymmetric Gaussian window is proposed to diminish the relevance of voice samples preceding GCIs on the WLP model, thus providing a quasi closed phase inverse filtering method. Characterization of symmetric and asymmetric GLP methods for glottal source estimation is addressed based on synthetic and natural phonation data, resulting in a set of optimal parameters for the Gaussian attenuation windows. The results show that the proposed asymmetric attenuation improves voice inverse filtering with respect to the symmetric GLP method. Comparisons with other state-of-the-art techniques suggest that the proposed GLP approaches are competitive, falling slightly short in performance only when contrasted with the well-known quasi closed inverse filtering analysis. The simplicity of implementing the attenuation windows, coupled with their robust performance, positions the proposed GLP methods as two attractive and straightforward voice inverse filtering techniques for practical application.

Validating the efficacy and value proposition of mental fitness vocal biomarkers in a psychiatric population: prospective cohort study.

The utility of vocal biomarkers for mental health assessment has gained increasing attention. This study aims to further this line of research by introducing a novel vocal scoring system designed to provide mental fitness tracking insights to users in real-world settings. A prospective cohort study with 104 outpatient psychiatric participants was conducted to validate the "Mental Fitness Vocal Biomarker" (MFVB) score. The MFVB score was derived from eight vocal features, selected based on literature review. Participants' mental health symptom severity was assessed using the M3 Checklist, which serves as a transdiagnostic tool for measuring depression, anxiety, post-traumatic stress disorder, and bipolar symptoms. The MFVB demonstrated an ability to stratify individuals by their risk of elevated mental health symptom severity. Continuous observation enhanced the MFVB's efficacy, with risk ratios improving from 1.53 (1.09-2.14, p=0.0138) for single 30-second voice samples to 2.00 (1.21-3.30, p=0.0068) for data aggregated over two weeks. A higher risk ratio of 8.50 (2.31-31.25, p=0.0013) was observed in participants who used the MFVB 5-6 times per week, underscoring the utility of frequent and continuous observation. Participant feedback confirmed the user-friendliness of the application and its perceived benefits. The MFVB is a promising tool for objective mental health tracking in real-world conditions, with potential to be a cost-effective, scalable, and privacy-preserving adjunct to traditional psychiatric assessments. User feedback suggests that vocal biomarkers can offer personalized insights and support clinical therapy and other beneficial activities that are associated with improved mental health risks and outcomes.

Acoustic feature analysis of directional sound from parametric array speakers with application to mitigate attacks on voice-activated systems

A number of studies have identified the vulnerabilities of voice-activated devices to attacks from amplitude-demodulated sound from directional speakers. In this research, a controlled set of speech data, both live and played back from directional speakers, is analyzed to identify the acoustic features integral to speaker recognition in smart devices. A category of features, including Mel-frequency cepstral coefficients derived from Mel filter banks, are examined for potential to differentiate and classify speech sources. Potential vulnerabilities in voice-activated systems are systematically identified by examining the acoustic characteristics of diverse voice samples. This analytical approach aims to unveil potential weaknesses that could be exploited, triggering responses from these systems. The findings of this study contribute to the comprehension of the security landscape surrounding voice-activated technologies, paving the way for enhanced cybersecurity measures within this domain.

Nonlinear Dynamic Analysis and Perturbation Measurement Used for Discriminating Pathological Voices and Their Correlations With Perceptual Evaluation

1. To investigate the discriminatory and diagnostic power of nonlinear dynamic analysis measures concerning voices from normal, benign, and malignant voice disorders. 2. To study the correlations of nonlinear dynamic analysis measures with perceptual ratings to evaluate the reliability of the objective acoustic analysis in predicting severity of voice. The perturbation analysis metrics used were Jitter%, Shimmer%, and signal-to-noise ratio. The nonlinear dynamic analysis metrics used were spectrum convergence ratio (SCR), nonlinear energy difference ratio (NEDR), and rate of divergence (ROD). Subjects were enrolled and divided into three groups based on laryngeal pathology: normal, benign, and malignant. Vowel sound and reading samples were recorded. Perceptual evaluation was applied to these voice samples to investigate correlations between metrics and auditory perception. Each metric was capable of discriminating laryngeal pathology, except for SCR in the case of distinguishing between benign and malignant pathologies. Perturbation analysis parameters had a moderate ability to differentiate between normal and benign pathologies, but were unable to characterize malignant pathologies for certain diseases, such as Reinke's edema. All metrics significantly correlated with perceptual G scores. Nonlinear dynamic analysis was superior when applied to cases of severe dysphonia, where linear metrics such as Jitter% and Shimmer% tended to lose utility. NEDR and ROD were successful at differentiating between the different pathologies, whereas SCR could not discriminate between the benign and malignant groups. Perturbation and nonlinear dynamic analyses are comparable in their discriminating power with respect to normal and benign voices, and normal and malignant voices. The nonlinear dynamic analysis metrics NEDR and ROD may be superior in clinical settings with respect to discriminating voice pathology ranging from mild pathological voice to severe dysphonia, and with respect to discriminating benign and malignant voice. SCR was found unable to discriminate pathological voices.