Voice Patterns Research Articles

History in the making: Voicing alternation as stop lenition via an automatic analysis of large-scale corpora in French and Spanish

This study focuses on voicing alternation in French and Spanish stops, i.e. canonically voiceless /ptk/ realized as voiced [bdg] or canonically voiced /bdg/ realized as voiceless [ptk]. Forced alignment with voicing variants was used to annotate large speech corpora in French and in Spanish. The following factors of variation were examined: position in the word, preceding and following context, duration of the stop and that of surrounding phones, speech rate, part of speech, and the weight of these factors on voicing alternation. The voicing nature of the stops (whether the stop is phonologically voiceless /ptk/ or voiced /bdg/) turns out to be the factor that contributes the most to the prediction of voicing alternations among all investigated factors for Spanish, according to the random forest model. Whereas for French, the same factor comes after contextual and acoustic factors in the ranking. These results suggest that stop voicing patterns differently in these two Romance languages, although they both have a similar voiced-voiceless phonological contrast.

Sentiment Predictor for Stress Detection Using Voice Stress Analysis

Abstract: This project proposes a machine learning basedapproach to predict the sentiment i.e Stressed, Unstressed orNeutral from voice of a human being and also presents a comparative analysis among various machine learning models. This technology seeks to distinguish between stressed and non-stressed outputs in response to stimuli. The dataset used fortraining in this project comprises audio recordings based onfour different situations, each recorded in various emotionalstates. Given the increased demand for communication betweenintelligent systems and human, automatic stress detection isbecoming an interesting research topic. Even while the level ofspecific hormones, such cortisol, can be precisely measured to indicate stress, this approach is not feasible for diagnosing stress in interactions between humans and machines. The Sentiment Predictor for Stress Detection Using Voice Stress Analysis utilizesmachine learning algorithms such as SVM, Random Forest , Logistic Regression etc. to analyze voice patterns and predict an individual’s emotional state based on the changes in their vocal characteristics. The technology can identify changes in pitch, tone, and rhythm, among other factors, to determinean individual’s stress level. Overall, the Sentiment Predictorfor Stress Detection Using Voice Stress Analysis is a promising technology that can provide valuable insights into an individual’s emotional state and support their well-being in various settings

Cross-Regional Patterns of Obstruent Voicing and Gemination: The Case of Roman and Veneto Italian

Italian has a length contrast in its series of voiced and voiceless obstruents while also presenting phonetic differences across regional varieties. Northern varieties of the language, including Veneto Italian (VI), are described as maintaining the voicing contrast but, in some cases, not the length contrast. In central and southern varieties, the opposite trend may occur. For instance, Roman Italian (RI) is reported to optionally pre-voice intervocalic voiceless singleton obstruents whilst also maintaining the length contrast for this consonant class. This study looks at the acoustic realization of selected obstruents in VI and RI and investigates (a) prevoicing patterns and (b) the effects and interactions of regional variety, gemination, and (phonological and phonetic) voicing on consonant (C) and preceding-vowel (V) durations, as well as the ratio between the two (C/V), with a focus on that particular measure. An acoustic phonetic analysis is conducted on 3703 tokens from six speakers from each variety, producing eight repetitions of 40 real CV́C(C)V and CVC(C)V́CV words embedded in carrier sentences, with /p, pp, t, tt, k, kk, b, bb, d, dd, ɡ, ɡɡ, f, ff, v, vv, t∫, tt∫, dʒ, ddʒ/ as the target intervocalic consonants. The results show that both VI and RI speakers produce geminates, yielding high C/V ratios in both varieties, although there are cross-regional differences in the realization of singletons. On the one hand, RI speakers tend to pre-voice voiceless singletons and produce overall shorter C durations and lower C/V ratios for these consonants. On the other hand, VI speakers produce longer C durations and higher C/V ratios for all voiceless singletons, triggering some overlap between the C length categories, which results in partial degemination through singleton lengthening, although only for voiceless obstruents. The implications of a trading relationship between phonetic voicing and duration of obstruents in Italian gemination are discussed.

From Manipulative Masterminds to Flirtatious Fools: An Examination of Male Sweet Voice Archetypes in Anime

Anime characters make use of many linguistic features, such as grammatical patterns, vocabulary, and vocal quality, to index specific character traits or archetypes. This includes both “role language”, associated with membership in social groups, and “character language”, associated with more abstract traits such as personality. This paper aims to expand upon the description of the archetypes conveyed through a traditionally feminine vocal quality called “sweet voice” and discusses how the use of this style in anime has broadened to encompass more masculine characters and archetypes as well. While women using sweet voice tend to be considered mature and genuine, interpretation of male sweet voice can range from gentlemanly, to morally questionable and manipulative, to something closer to flirtatious comic relief. Male sweet voice characters, similar to their female counterparts, have certain patterns in their word choice and grammatical speaking styles, although these stylistic patterns are less universal for male characters. Not only are the male sweet voice archetypes in anime different from those of female sweet voice characters, but these male characters are also able to freely swap styles, in vocal quality, vocabulary and grammar, where female characters cannot. Furthermore, the use of female sweet voice patterns by male voice actors can be used specifically to code characters as transgender women, something not possible with female sweet voice. These differences demonstrate that male sweet voice is not just female sweet voice overlaid onto male characters, but rather a specific evolution of the female styles specific to male characters.

Linguistic Characteristics In The Research Background And Problem Statement In The Doctoral Thesis Of Arabic EFL Students In Malaysia

This study explores the linguistic features within the move structures of the research background and problem statement sections in Chapter One of PhD theses written by Arab students in Malaysia. The analysis focuses on four key language elements: verb tense, modality, voice (active and passive), and personal pronouns. A corpus of 30 theses was analyzed to determine common linguistic patterns, revealing significant insights into the language choices of these academic writers. The study employs concordance and hand-tagged analysis to examine verb tense and voice, with manual verification for modality and personal pronouns to ensure accuracy. Eight distinct verb tenses were identified, with the present simple tense being the most frequently used, representing 55.30% of the total instances. This indicates a preference for focusing on current knowledge and ongoing research. In contrast, the past simple and future simple tenses were used to describe past methodologies and future implications, respectively. Modal auxiliaries play a crucial role in conveying sentiments and opinions. The study identified seven modal verbs, with 'can' being the most commonly used, suggesting a focus on possibility over obligation. The analysis of voice patterns revealed a higher occurrence of active voice (79.66%), especially in moves that conveyed up-to-date knowledge. In contrast, passive voice was more common in methodological descriptions, indicating a focus on the actions or subjects rather than the actors. The study also examined the usage of personal pronouns, finding a higher frequency of third-person pronouns, suggesting a more objective narrative style. These findings offer valuable insights into the linguistic choices of Arab EFL students in academic writing, providing a deeper understanding of their communicative intentions and alignment with academic discourse conventions. The results can guide future research on linguistic patterns in academic writing and help educators support EFL students in improving their writing skills.

(Invited) Triboelectric Nanogenerators for Self-Powered Sensing and Human-Machine-Interfaces

Wearable and portable electronics are garnering substantial interest for applications ranging from personalized healthcare monitoring to implantable devices and prosthetics. The demand for long-term operation poses significant challenges to their power supply. To address this, the development of novel self-powered systems holds considerable promise. In this talk, I will discuss our recent advancements in crafting triboelectric nanogenerators (TENGs) for self-powered sensing and human-machine interfaces (HMIs). I will first introduce our multifunctional tactile sensors inspired by human skin. These sensors combine a triboelectric nanogenerator layer with a piezoresistive sensing layer, emulating the ability of human skin to detect a variety of static and dynamic stimuli. This sensor can successfully recognize voice patterns, monitor physiological signals, and track human motion in real-time. Then, I will unveil our self-powered smart packaging system, which is driven by a novel desiccant-based triboelectric nanogenerator (D-TENG). This system offers a viable method for continuous monitoring of product conditions and quality control throughout the supply chain, thus minimizing food waste and maintaining a comprehensive log of logistics, especially critical in vaccine transport. Lastly, I will present our novel breath-driven triboelectric sensor in respiratory monitoring. This sensor can precisely detect breath variations and convert different breathing patterns into electrical signals, all without impeding normal breathing. It enables severely disabled individuals to control household appliances through a breath-driven HMI and provides an intelligent monitoring system for emergency situations. This innovative work paves the way for future development in self-powered sensors and advanced HMIs, enhancing assistive technology for individuals with disabilities.

Enhancing speaker identification in criminal investigations through clusterization and rank-based scoring

This paper introduces an approach that supports speaker identification in criminal investigations, specifically addressing challenges associated with large volumes of audio recordings featuring unknown speaker identities. Our approach clusters related recordings – potentially from the same person – based on representative voice embeddings extracted using the ECAPA-TDNN speaker recognition model. Grouping audio recordings from the same person enhances variability and richness in voice patterns, thereby improving confidence in automatic speaker recognition. We propose a combination of cosine similarity and a rank-based adjustment function to determine matches of audio clusters with individuals in an enrollment database. Our approach was validated through experiments on a Common Voice-based synthesized dataset and a real-life application involving cell phones seized in prisons, which contained thousands of conversational audio recordings. Results demonstrated satisfactory performance and stability, consistently reducing the pool of candidate speakers for subsequent analysis by a human investigator.

Pathological voice classification using MEEL features and SVM-TabNet model

In clinical settings, early diagnosis and objective assessment depend on the detection of voice pathology. To classify anomalous voices, this work uses an approach that combines the SVM-TabNet fusion model with MEEL (Mel-Frequency Energy Line) features. Further, the dataset consists of 1037 speech files, including recordings from people with laryngocele and Vox senilis as well as from healthy persons. Additionally, the main goal is to create an efficient classification model that can differentiate between normal and abnormal voice patterns. Modern techniques frequently lack the accuracy required for a precise diagnosis, which highlights the need for novel strategies. The suggested approach uses an SVM-TabNet fusion model for classification after feature extraction using MEEL characteristics. MEEL features provide extensive information for categorization by capturing complex patterns in audio transmissions. Moreover, by combining the advantages of SVM and TabNet models, classification performance is improved. Moreover, testing the model on test data yields remarkable results: 99.7 % accuracy, 0.992 F1 score, 0.996 precision, and 0.995 recall. Additional testing on additional datasets reliably validates outstanding performance, with 99.4 % accuracy, 0.99 F1 score, 0.998 precision, and 0.989 % recall. Furthermore, using the Saarbruecken Voice Database (SVD), the suggested methodology achieves an impressive accuracy of 99.97 %, demonstrating its durability and generalizability across many datasets. Overall, this work shows how the SVM-TabNet fusion model with MEEL characteristics may be used to accurately and consistently classify diseased voices, providing encouraging opportunities for clinical diagnosis and therapy tracking.

Implementasi Metode Jaringan Saraf Tiruan Backpropagation Pada Pengenalan Suara Manusia

Speech recognition is a process of voice identification using specific parameters taken by the sound catcher. The development of technology gave rise to an event that requires a calculation model on a computer system in speech recognition to be useful in science. One of the computer systems is the Backpropagation Artificial Neural Network (JST). This research uses the Backpropagation method in human speech recognition with the aim of knowing the architecture model and the level of accuracy obtained. Linear Predictive Coding (LPC) is used for voice feature extraction. Voice features in the time domain are converted into the frequency domain using Fast Fourier Transform (FFT). The voice data was divided into 80% training data and 20% testing data. A suitable JST architecture model is selected through training by calculating the optimal weights and biases to recognize the voice patterns well. The best architecture model found was 64-15-1-1. The model was tested using test data to test its ability to recognize voice patterns. Evaluation was done using K-Fold Cross Validation to measure the accuracy of the model. The accuracy value against the training data is 0.95, while against the testing data is 0.088886. The JST architecture model is very good at recognizing voices in training data, but less good in testing. Hopefully, this method can help in the research process related to recognition.

Regional differences (or lack thereof) in rendaku in Japanese surnames

No study has thoroughly investigated the regional differences of Japanese compound voicing known as rendaku. The present study addresses this issue by conducting a large-scale web-based survey with 492 Japanese speakers as participants and 1,776 compound surnames as stimuli. The results show no clear effects of dialects on rendaku application. This raises a novel theoretical issue for further investigation: Even though pitch accent has been argued to be inversely associated with rendaku (e.g., Sugito 1965; Zamma & Asai 2017), dialects with a variety of accentuation patterns nonetheless exhibit very similar voicing patterns in compound surnames. A tentative account based on foot structure is proposed.

AI-driven biometrics for secure fintech: Pioneering safety and trust

AI-driven biometrics is revolutionizing the security landscape in the financial technology (FinTech) sector, enhancing safety measures and fostering trust among users. This review explores the role of AI-driven biometrics in securing FinTech operations, highlighting its benefits and implications. AI-driven biometrics refers to the use of artificial intelligence (AI) algorithms to analyze biological data for authentication purposes. This technology has gained significant traction in the FinTech sector due to its ability to provide a higher level of security than traditional authentication methods. By analyzing unique biological traits such as fingerprints, facial features, and voice patterns, AI-driven biometrics can accurately verify the identity of users, making it difficult for unauthorized individuals to gain access to sensitive financial information. One of the key benefits of AI-driven biometrics in FinTech is its ability to enhance security measures. Traditional authentication methods such as passwords and PINs are increasingly vulnerable to hacking and fraud. AI-driven biometrics, on the other hand, offers a higher level of security by using biological traits that are unique to each individual. This makes it significantly more difficult for fraudsters to gain access to sensitive financial information. In addition to enhancing security, AI-driven biometrics also improves the user experience. By replacing traditional authentication methods with biometric authentication, users can access their accounts more quickly and conveniently, without the need to remember complex passwords or PINs. This not only improves the overall user experience but also reduces the risk of user error and account lockouts. Overall, AI-driven biometrics is pioneering safety and trust in the FinTech sector by providing a higher level of security and enhancing the user experience. As this technology continues to evolve, it is likely to play an increasingly important role in securing financial transactions and fostering trust among users.

Officers and men and fallen heroes: The discursive construction of regimented masculinity in the Nigerian Army

This paper examines the use of gendered language in the Nigerian Army's community of practice through the application of insights from language ideology and theory of masculinity. Data were generated by means of participant observation and semi-structured interviews in a one-year fieldwork involving representative sample of 18 personnel of the 6 Battalion, Ibawa and 2 Brigade, Uyo, both in Akwa Ibom State in South-eastern Nigeria. The findings show that the Nigerian Army maintains institutional gendered language practices used among its personnel in regimented functions and social events. The gendered registers occur in the soldiers' generic use of male address terms, adaptation to male-coded voice pattern in parades, masculinisation of Army's workout songs, and the subordination of femininity in institutional associations, all combined to construct the regimented and performative masculinisation of the profession. This practice is observed to be informed by the numerical domination of men in the profession that was originally perceived as males'; a conception that has shaped the linguistic ideology and performance of the Nigerian Army to rehearse masculine orientations. It is however recommended that the Army's language practices should capture modern ideals of a gender sensitive world that connect to the clamour for gender equality and equal social belonging through the inclusion of feminine linguistic markers in workplaces.

Development of a non-invasive Covid-19 detection framework using explainable AI and data augmentation1

This paper investigates the potential of COVID-19 detection using cough, breathing, and voice patterns. Speech-based features, such as MFCC, zero crossing rate, spectral centroid, spectral bandwidth, and chroma STFT are extracted from audio recordings and evaluated for their effectiveness in identifying COVID-19 cases from Coswara dataset. The explainable AI SHAP tool is employed which identified MFCC, zero crossing rate, and spectral bandwidth as the most influential features. Data augmentation techniques like random sampling, SMOTE, Tomek, and Edited Nearest Neighbours (ENN), are applied to improve the performance of various machine learning models used viz. Naive Bayes, K-nearest neighbours, support vector machines, XGBoost, and Random Forest. Selecting the top 20 features achieves an accuracy of 73%, a precision of 74%, a recall of 94%, and an F1-score of 83% using the Random Forest model with the Tomek sampling technique. These findings demonstrate that a carefully selected subset of features can achieve comparable performance to the entire feature set while maintaining a high recall rate. The success of the Tomek undersampling technique highlights the ability of model to handle sparse clinical data and predict COVID-19 and associated diseases using speech-based features.

Psychological Health Assessment and Intervention System for Chemical Engineering Students Based on Emotional Computing Technology

Emotional computing technology is revolutionizing health assessment by integrating emotional and psychological factors into traditional diagnostic processes. This innovative approach utilizes advanced algorithms and sensor technologies to analyze individuals' emotional states and mental well-being, providing valuable insights into their overall health. By detecting subtle changes in facial expressions, voice patterns, and physiological signals, emotional computing technology can assess stress levels, mood fluctuations, and mental health conditions with greater accuracy and sensitivity. This paper presents a novel Psychological Health Assessment and Intervention System tailored for chemical engineering students, leveraging Emotional Computing Technology enhanced by Cartesian Feature Weighted Emotional Classification (CFWEC). The system aims to assess students' psychological well-being by analyzing emotional cues extracted from various sources such as facial expressions, voice patterns, and text interactions. Through simulated experiments and empirical validations, the efficacy of CFWEC-enhanced emotional computing technology is evaluated in predicting and intervening in psychological health issues among chemical engineering students. Results demonstrate significant improvements in accuracy and sensitivity compared to traditional assessment methods. For instance, the CFWEC model achieved an average accuracy rate of 85% in identifying students at risk of psychological distress, enabling timely interventions. Additionally, the system provides personalized recommendations and interventions based on individual emotional profiles, leading to improved mental health outcomes. These findings underscore the potential of Emotional Computing Technology with CFWEC in promoting psychological health and well-being among chemical engineering students.

Detection of Synthetically Generated Speech By IJISRT

Deepfake technology has made it increasingly difficult to discern real from fabricated audio, posing a significant challenge in the digital age. By employing sophisticated algorithms and voice recognition techniques, the system proposed in this article can analyse voice patterns and nuances to spot inconsistencies and anomalies, which are common indicators of deepfake voices and prevent scams and other types of cyber security issues.

Examining city image construction through the lens of linguistic soundscape: a case analysis of Shenzhen

The study examined the role of linguistic soundscape in constructing city image. Previous studies paid inadequate attention to linguistic soundscape, an important branch of linguistic landscape. The authors applied the theory of City Identity System (CIS) and selected Shenzhen, the third wealthiest city in China, to discuss how linguistic soundscape in public space may shape people’s perceptions of a city. The public address announcements in the major metro stations, airport, and ferry terminal of Shenzhen were recorded and analysed based on their language pattern, functions, and gender of voice. It was concluded that despite issues such as an imbalance between informative and symbolic functions, the linguistic soundscape of Shenzhen generally exhibited positive influence and promising potential in the construction of city image.

Real Time Bimodal Emotion Recognition using Hybridized Deep Learning Techniques

Emotion detection has become an important study subject due to the potential uses that it has in many other domains. It incorporates data from various sources, such as facial expressions, voice patterns, physiological indications, and textual signals. To identify human emotions, a real-time bimodal image and audio-based detection-based approach has been proposed. The approach is termed as “bimodal” since images and audio data are considered as inputs to the proposed model. In this study, hybridized CNN-LSTM for the image-based model and Multi-Layer Perceptron is used for the developing audio model. For the image-based model, two combined layers of CNN, and LSTM along with regularization techniques like drop-out, and batch normalization have been adopted for model development. Similarly, for the audio model, the dataset has been fine-tuned using a standard scaler and PCA before training. Based on the combined results of both models, the user’s final emotion is determined. Both the audio-based model and the video-based model outperform the existing models discussed in the literature with higher performance accuracy of 79.34% and 77.27% respectively.

On the impact of downward-directed human voice radiation on ground reflections

Previous studies on vertical or full-spherical directivity patterns of the human voice showed that the human voice has a slightly downward main radiation direction over a wide frequency range. This paper investigates the phoneme-dependencies of human voice radiation in the vertical plane and analyzes to what extent these characteristics affect the ground reflection and the sound incidence at a listener position. The results show that for most phonemes and below 800 Hz, the ground reflection is stronger than the direct sound component because of the downward-directed main radiation. On the contrary, between 800 Hz and 1.6 kHz, the main radiation direction is upward, probably mainly due to diffraction and reflections from the shoulders and the torso.

Did the speech of patients with Myasthenia Gravis decline over 4 years?

To compare the speech and voice patterns of myasthenia gravis (MG) patients over four years, and correlate the results with clinical aspects of the disease. Data was collected for 4 years. The clinical assessment tools included the Quantitative Myasthenia Gravis (QMG) score, the Myasthenia Gravis Foundation of America (MGFA) clinical classification, and the Myasthenia Gravis Quality of Life 15-item Scale (MG-QoL). To assess speech, the recorded speaking tasks were analyzed acoustically and given auditory-perceptual ratings. Sex (equal distribution) and age (p=0.949) were used as matching criteria in the final sample, which consisted of 10 individuals in the MG group (MGG) and 10 individuals in the control group (CG). After 4 years, the MG participants presented stable health status, increased mild and moderate dysarthria (from 40% to 90% of the subjects), and a significant deterioration in the respiration, phonation, and articulation subsystems. The acoustic analysis showed a decline in articulatory patterns (speech rate p=0.047, articulation rate p=0.007, mean syllable duration p=0.007) and vocal quality (increased jitter p=0.022). In the follow-up comparison, there was a significant difference between the phonation variables (shimmer and harmonic-to-noise ratio) of the MGG and CG. The MG patients presented a decline in speech over four years and an increase in mild and moderate dysarthria. Despite presenting stable health status, their respiratory, phonatory, and articulatory subsystems worsened. There was no correlation between speech patterns and clinical characteristics of the disease (severity and motor scale).

Mel Frequency Cepstral Coefficients (MFCC) Method and Multiple Adaline Neural Network Model for Speaker Identification

Speech recognition technology makes human contact with the computer more accessible. There are two phases in the speaker recognition process: capturing or extracting voice features and identifying the speaker's voice pattern based on the voice characteristics of each speaker. Speakers consist of men and women. Their voices are recorded and stored in a computer database. Mel Frequency Cepstrum Coefficients (MFCC) are used at the voice extraction stage with a characteristic coefficient of 13. MFCC is based on variations in the response of the human ear's critical range to frequencies (linear and logarithmic). The sound frame is converted to Mel frequency and processed with several triangular filters to get the cepstrum coefficient. Meanwhile, at the speech pattern recognition stage, the speaker uses an artificial neural network (ANN) Madaline model (many Adaline/ which is the plural form of Adaline) to compare the test sound characteristics. The training voice's features have been inputted as training data. The Madaline Neural Network training is BFGS Quasi-Newton Backpropagation with a goal parameter of 0,0001. The results obtained from the study prove that the Madaline model of artificial neural networks is not recommended for identification research. The results showed that the database's speech recognition rate reached 61% for ten tests. The test outside the database was rejected by only 14%, and 84% refused testing outside the database with different words from the training data. The results of this model can be used as a reference for creating an Android-based real-time system.