Speech Processing

NLP research on aligning lexical representation spaces to one another has so far focused on aligning language spaces in their entirety. However, cognitive science has long focused on a local perspective, investigating whether translation equivalents truly share the same meaning or the extent that cultural and regional influences result in meaning variations. With recent technological advances and the increasing amounts of available data, the longstanding question of cross-lingual lexical alignment can now be approached in a more data-driven manner. However, developing metrics for the task requires some methodology for comparing metric efficacy. We address this gap and present a methodology for analyzing both synthetic validations and a novel naturalistic validation using lexical gaps in the kinship domain. We further propose new metrics, hitherto unexplored on this task, based on contextualized embeddings. Our analysis spans 16 diverse languages, demonstrating that there is substantial room for improvement with the use of newer language models. Our research paves the way for more accurate and nuanced cross-lingual lexical alignment methodologies and evaluation.

Speech is the expression of or the ability to express thoughts and feelings by articulate sounds. It is the main way of communication between humans. There are thousands of languages used in the world. Speech recognition is a process of recognition of human speech by computer and giving the string output of spoken sentence in written form. There are lots of advantages of speech recognition. There are many methods of speech recognition but yet we have not get 100% result of speech recognition. Here in this paper we will explain the development in speech recognition from 1952 to 2014. Finally we will give conclusion that which approach to speech recognition is best and will be beneficial for the future in the field of speech recognition. Keywords—Automatic Speech Recognition (ASR), AcousticPhonetic approach, pattern-comparison technique, Artificial Intelligence approach, Hidden Markov Model (HMM).

The paper discusses aspects of data research, in-depth data analysis, knowledge acquisition, methods of data processing in the knowledge base, methods of intellectual analysis, and application of data mining in the field of medicine. A group of HIV-infected patients was identified, an analysis with a medical history was carried out, models and an algorithm of actions (input data) were developed, and analysis and experiments with data search methods were carried out. All diseases were presented as a set of numerical vectors and were grouped into clusters, according to the described methods, and with the help of this distribution, the Hopkins statistics value was calculated. Clustering itself was carried out using the usual tools of the sklearn library. Various methods of representation of multidimensional data in a two-dimensional plane are proposed, such as the method of basic components, the Kohonen line, etc. Two different clustering methods were considered, namely, the k-Medium method (using the Kmeans function from the Python sklearn library) and densitybased clustering methods with autoconfiguration (from the HDBSCAN function from the Python Hdbscan library). In the case of comparison, the cluster structure is evaluated by changing various parameters of one algorithm (for example, the number of k groups); a model (or several) is built on the received and prepared objects, and its parameters are adjusted. After that testing and analysis of the results were carried out.

We present the Ultrasonic Consciousness Hypothesis, proposing that the systematic removal of ultrasonic frequencies (20-96kHz) through lossy audio compression since the 1990s may have inadvertently eliminated crucial emotional grounding mechanisms in human consciousness. Through empirical analysis of 192kHz vinyl recordings revealing 58,416 temporal resonances with Fibonacci correlations of 0.85-0.88, combined with the Marine Algorithm's O(1) jitter-based salience detection, we identify three distinct types of spectral fractures: dynamic fractures from loudness normalization, spectral fractures from frequency truncation, and perceptual fractures from engineered emotional responses. We demonstrate that frequencies beyond conscious auditory perception may serve as subliminal carriers of emotional resolution patterns, with their removal potentially contributing to observed increases in anxiety, depression, and social polarization. This paper introduces a theoretical framework for understanding audio as a complete consciousness substrate, proposes quantitative metrics for measuring spectral completeness, and presents preliminary evidence suggesting that the "vinyl resurgence" may represent an unconscious societal attempt to restore full-spectrum emotional nutrition. While correlational rather than causal, our findings suggest that two decades of compressed audio consumption represents an unprecedented experiment in human consciousness modification.

In recent years, speech recognition technology has become a more common notion. Speech quality and intelligibility are critical for the convenience and accuracy of information transmission in speech recognition. The speech processing systems used to converse or store speech are usually designed for an environment without any background noise. However, in a real-world atmosphere, background intervention in the form of background noise and channel noise drastically reduces the performance of speech recognition systems, resulting in imprecise information transfer and exhausting the listener. When communication systems’ input or output signals are affected by noise, speech enhancement techniques try to improve their performance. To ensure the correctness of the text produced from speech, it is necessary to reduce the external noises involved in the speech audio. Reducing the external noise in audio is difficult as the speech can be of single, continuous or spontaneous words. In automati...

Text processing in Serbian is based on the Intex format system of electronic dictionaries. Although lexical recognition is successful for 75% to 90% of word forms (depending on the type of text), some categories of words remain unrecognized. In this paper we present two aspects of e-dictionary enhancement that provide for additional recognition of two important categories of words: named entities and words generally not recorded in traditional dictionaries. We first describe the structure and content of dictionaries of proper names, both personal and geographic, developed to recognize the corresponding classes of named entities. Then we present a set of lexical transducers expressing morphological rules governing word formation, developed for the recognition of unknown words. The resources presented significantly improve the lexical recognition process.

The world economy is evolving at a very fast rate with revolutionary technologies altering the conventional patterns of business and governance. Among them, blockchain and big data are two prominent forces that are transforming the way transactions, information, and policies are managed. Blockchain offers a decentralized and immutable means of storing information, while big data facilitates analysis on an unprecedented scale. These technologies are also driven by government policies that encourage or control their development. This essay analyzes the structure of blockchain, the influence of big data, and the government role in encouraging or controlling their usage. Based on an analysis of empirical evidence and existing global tendencies, this research intends to present an integrated view of the interrelation among these three drivers and how they drive the world digital economy. As more nations embrace digital infrastructure, their use of blockchain and big data becomes increasingly prevalent in private and public institutions. Companies are applying the technologies to reduce costs, enhance transparency, and enhance customer experience. Governments are also trying to come up with regulations that facilitate ethical and safe usage of digital resources. The report also discusses how the rest of the world is catching up on technology trends through innovation and policy interventions. Issues like data protection, cybersecurity, and environmental effects of data processing are identified. The report concludes that multi stakeholder cooperation between governments, industry, and technology stakeholders is imperative in establishing an open and secure digital future.

Oral proficiency testing plays a critical role in language assessment; however, classic ASR faces such problems as Americans Bias, which means difficulty for ASR as application to recognize non-American accents, and the evaluation inaccuracy. The major problem here is that classic ASR systems cannot recognize and assess non-American accents as they would reward non-biased, unbiased, and fair scores for speakers with regional or non-native English accents. This bias in particular distorts the fairness of the oral proficiency test. The current approaches are inefficient with regards to the fairness and standardization in natural language scenarios. Unfortunately, in practice the conventional ASR models are not perfect and become a source of errors and biases when trying to assign a score for the given speech; the specific problem with the current approach is that the speakers with regional or nonnative accent will be almost impossible to score correctly. To address this bias, this study proposes BERTAdv, incorporating BERT which focuses on contextual understanding into the Adversarial training method for enhancing the ASR's accuracy in OPI. In an attempt to improve the BERTAdv to analyze both simple and complex linguistic attributes, the model is trained on a large and balanced dataset. The model presents desirable accuracy with a high R-squared of 0.93, MAE of 0.26, and RMSE of 0.46 and is more accurate and fairer than existing methods.

This paper describes a method to use the thresholding technique to automatically classify the three parts of a speech signal - silence, unvoiced and voiced. It makes use of three characteristics of the speech signal namely - short time energy (STE), zero crossing rate (ZCR) and fundamental frequency (F0).

With the help of automatic speech recognition (ASR) techniques, computers become capable of recognizing speech. The Quran is the speech of Allah (The God); it is the Holy book for all Muslims in the world; it is written and recited in Classical Arabic language, the language in which it was revealed by Allah to the Prophet Muhammad. Knowing how to pronounce correctly the Quranic sounds and correct mistakes occurred in reading is one of the most important topics in Quranic ASR applications, which assist self-learning, memorizing and checking the Holy Quran recitations. This paper presents a practical framework for development and implementation of an optimal ASR system for Quranic sounds recognition. The system uses the statistical approach of Hidden Markov Models (HMMs) for modeling the Quranic sounds and the Cambridge HTK tools as a development environment. Since sounds duration is regarded as a distinguishing factor in Quranic recitation and discrimination between certain Quranic sounds relies heavily on their durations, we have proposed and tested various strategies for modeling the Quranic sounds' durations in order to increase the ability in distinguishing them properly and

This paper presents the TYPALOC corpus of French Dysarthric and Healthy speech and the rationale underlying its constitution. The objective is to compare phonetic variation in the speech of dysarthric vs. healthy speakers in different speech conditions (read and unprepared speech). More precisely, we aim to compare the extent, types and location of phonetic variation within these different populations and speech conditions. The TYPALOC corpus is constituted of a selection of 28 dysarthric patients (three different pathologies) and of 12 healthy control speakers recorded while reading the same text and in a more natural continuous speech condition. Each audio signal has been segmented into Inter-Pausal Units. Then, the corpus has been manually transcribed and automatically aligned. The alignment has been corrected by an expert phonetician. Moreover, the corpus benefits from an automatic syllabification and an Automatic Detection of Acoustic Phone-Based Anomalies. Finally, in order to...

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Best Tree Encoding (BTE) is a promising feature extraction technique based on wavelet packet decomposition that is utilized in Automatic Speech Recognition (ASR). This research introduces an enhancement of Wavelet Packet Best Tree (WPBT) Calculations. The standard features BTE encodes the tree structure using a mathematical model into a features vector of 4 components. The best tree structure has been calculated using the entropy function. In the standard version of BTE, Shannon entropy has been chosen as the entropy function. In this research, Shannon Entropy (SE), Renyi Entropy (RE), and Tsallis Entropy (TE) are used to construct the Best Tree. The encoding of the Best Tree has been done using the same mathematical model approach in the standard 4-Point BTE. The proposed model is tested and Verified against the most widely used feature Mel Frequency Cepstral Coefficient (MFCC) plus delta and delta-delta coefficients (39 parameters) to evaluate its performance. The TIMIT database is used in this research. All phones are divided into five classes: Vowels, Fricatives, Silences, Nasals, and Plosives. The acoustical model has been implemented using Hidden Markov Model (HMM). No language model has been applied. The HMM Tool Kit (HTK) software is used for model implementation. The experiments show that BTE using Tsallis entropy yields the highest overall success rate of 75.85% which is better than MFCC's overall success rate of 71.76%. Comparing the vector of 4 components of BTE to the 39 components vector of MFCC makes it a very promising feature vector to be considered for research and development.

Under an ARC Linkage Infrastructure, Equipment and Facilities (LIEF) grant, speech science and technology experts from across Australia have joined forces to organise the recording of audio-visual (AV) speech data from representative speakers of Australian English in all capital cities and some regional centres. The Big Australian Speech Corpus (the Big ASC) will provide a standard recording setup and a collaboratively-designed elicitation protocol to create a corpus of AV speech data incorporating annotations and metadata, accessible via a centralised storage facility. The Big ASC infrastructure will provide a significant boost to research in speech science and human communication in Australia.

Parkinson’s  disease  (PD)  is  a  neurodegenerative  disorder  that  affects  the  coordination of muscles and limbs, including those responsible of the speech production. The lack of control of the limbs and muscles involved in the speech production  process  can  generate  intelligibility  problems  and  this  situation  has  a  negative  impact  in  the  social  interaction  of  the  patients.  It  is  already  demonstrated that constant speech therapy can improve the communication abilities of the patients; however, the measurement of the recovery progress is done subjectively by speech therapists and neurologists. Due to this, it is required the development of flexible tools able to asses and guide the speech therapy  of  the  patients.  In  this  paper  the  design  and  deployment  of  a  new  device  for  the  real  time  assessment  of  speech  signals  of  people  with  PD  is  presented. The processes of design and deployment include the development on three platforms: f...

Processing incoming sensory information and transforming this input into appropriate motor responses is a critical and ongoing aspect of our moment-to-moment interaction with the environment. While the neural mechanisms in the posterior parietal cortex (PPC) that support the transformation of sensory inputs into simple eye or limb movements has received a great deal of empirical attention—in part because these processes are easy to study in nonhuman primates—little work has been done on sensory-motor transformations in the domain of speech. Here we used functional magnetic resonance imaging and multivariate analysis techniques to demonstrate that a region of the planum temporale (Spt) shows distinct spatial activation patterns during sensory and motor aspects of a speech task. This result suggests that just as the PPC supports sensorimotor integration for eye and limb movements, area Spt forms part of a sensory-motor integration circuit for the vocal tract.

This paper introduces a new learning algorithm for human activity recognition capable of simultaneous regression and classification. Building upon Conditional Restricted Boltzmann Machines (CRBMs), Factored Four Way Conditional Restricted Boltzmann Machines (FFW-CRBMs) incorporate a new label layer and four-way interactions among the neurons from the different layers. The additional layer gives the classification nodes a similar strong multiplicative effect compared to the other layers, and avoids that the classification neurons are overwhelmed by the (much larger set of) other neurons. This makes FFW-CRBMs capable of performing activity recognition, prediction and self auto evaluation of classification within one unified framework. As a second contribution, Sequential Markov chain Contrastive Divergence (SMcCD) is introduced. SMcCD modifies Contrastive Divergence to compensate for the extra complexity of FFW-CRBMs during training. Two sets of experiments one on benchmark datasets and one a robotic platform for smart companions show the effectiveness of FFW-CRBMs.

The pervasive development disorders in autism condition lead to impairments in language and social communication. They are evidenced as atypical prosody production, emotion recognition and apraxia, among others communication deficits. This work tackle with the problem of the recognition of pathologies derived from these disorders in children, based on the acoustic analysis of speech. Specifically, the task consists of the diagnosis of normality (typically developing children) or three different pathologies. We propose an evolutionary approach to the feature selection stage. It relies on the use of genetic algorithm to find the set of features that optimally represent the speech data for this classification task. The genetic algorithm uses a support vector machine in order to evaluate the solutions (each individual) during the search. The results showed that our methodology improves the baseline provided for the task. The obtained unweighted classification accuracy was 54.80% on the development set, which represents a relative improvement of 6%, and 55.41% on test set. On the related task of binary classification between typical versus atypical developing condition, our approach achieved an unweighted classification accuracy of 92.66% on the test set. Index Terms-evolutionary wrapper, speech feature selection, autism disorders The dataset used consists of three sets of the Child Pathological Speech Database (CPDS): train, development and test sinc(i) Research Center for Signals, Systems and Computational Intelligence (fich.unl.edu.ar/sinc)

Although images as viewed from intermediate virtual viewpoints can be synthesized using texture and depth maps from nearby camera views via depth-image-based rendering (DIBR), the rendered images contain disocclusion holesspatial regions that were not visible in the reference views due to foreground object occlusion-that requires proper filling. In this paper, we introduce a new signal prior into the hole filling problem formulation: given disocclusion holes are part of the background and background tends to have low visual saliency, the extrapolated signal into the holes must also be of low saliency. Mathematically, we add a low-saliency prior to an exemplar-based inpainting algorithm, so that the best-matched block has both small matching cost and is of low visual saliency. Moreover, we compute a suitable Lagrange multiplier value for the saliency cost term via analysis of the reference images. Experimental results show that using a low-saliency prior can improve performance by 0.5 dB over a previous hole filling scheme.

We approach the problem of understanding the disease progression and speech of patients suffering from ALS as a language divergence identification problem. We summarize the promises and challenges of using speech-related biomarkers to adapt speech recognition to ALS patients and correlate language divergence with disease progression.

Streszczenie. w artykule omówiono własności najczęściej wykorzystywanych masek filtrów krawędziowych (operatorów kontekstowych) laplace'a stopnia trzeciego i piątego wraz ze schematami różnicowymi, które stanowią podstawę ich wyprowadzenia. Do konstrukcji wykorzystano metody różnic skończonych i elementu skończonego (mes) z aproksymacją rozwiązania funkcjami biliniowymi (przestrzeń lagrange'a elementu skończonego). zaproponowano nowe maski konwolucyjne indukowane przez schematy różnicowe operatora laplace'a. Każdej z omawianych dziesięciu masek przypisano tzw. Π-formę pierwszego przybliżenia różniczkowego. na jej podstawie można określić rząd schematu różnicowego aproksymującego operator ∇ 2 , a tym samym -rząd maski (rząd maski i stopień maski są różnymi wskaźnikami). Ponadto (i to jest najważniejsze), można jednoznacznie stwierdzić, czy dana maska jest rzeczywiście maską laplace'a. w pracy wyjaśniono matematyczne podstawy i pochodzenie kilku stosowanych w praktyce filtrów laplace'a oraz zwrócono uwagę na pewne nieścisłości (powielane w literaturze) pojawiające się w ich opisach dyskretnych. ich konsekwencje przedstawiono na kilku wybranych zdjęciach satelitarnych pól zachmurzenia zawierających rozbudowaną chmurę Cumulonimbus oraz na wygenerowanym w pakiecie matlab  fragmencie grafiki dwu-i trójwymiarowej. wskazano elementy, które obowiązkowo powinny być uwzględniane w procedurze porównywania własności masek filtrów liniowych. Praca ma charakter teoretyczny. Prowadzone tu badania na poziomie podstawowym odwołują się do kilku przykładów praktycznych, które pełnią funkcję ilustracji wyprowadzanych wniosków. zdajemy sobie sprawę z faktu, że jednoznaczne, a nawet kategoryczne sformułowania końcowe oraz wskazanie obszarów zastosowania wyników zawsze związane jest długotrwałymi doświadczeniami oraz z częstym upowszechnianiem rezultatów. Przedstawiamy zatem wyłącznie zwartą procedurę określania matematycznych własności masek filtrów krawędziowych laplace'a.

Text-to-Speech (TTS) synthesis is a problem almost as old as Natural Language Processing (NLP). The focus of this problem is on creating tools capable of generating a voice uttering a given text. Deep learning-powered solutions try to go beyond mere speech generation from the text: newer models try to factorise the probability predicted by these generative models to condition it on various aspects: speaker's voice, speaking style, or prosody. In this work, we focused on predicting the speaking style from the given text inside a conversation, for the application to conversational agents and chatbots. To this end, we developed and trained a neural network module working as a connector between the textual component of a chatbot (i.e., the neural language model for dialogue understanding and generation) and the speech synthesis component of a chatbot (i.e., the neural TTS synthesis model with speaking style conditioning).

Recently, a number of (approximate) approaches emerged in speech processing, which try to overcome the known lack of match between symbol level evaluation measures (e.g., word error rate) and the standard string (symbol sequence) cost (e.g., sentence error)-based Bayes decision rule, by using symbol level cost functions for Bayes decision rule. Nevertheless, experiments show that for a majority of test samples both decision rules still give equal decisions, especially at lower error rates. In this paper, analytic evidence for these observations is provided. A set of conditions is presented, for which Bayes decision rule with symbol level and string level cost function leads to the same decisions. Furthermore, the case of word error cost represented by the Levenshtein (edit) distance is investigated, which upon others covers the important case of speech recognition. A Hamming distance-based upper bound to the Levenshtein cost function is discussed. This cost function relates to former, word-posterior based decision rules, and the corresponding efficient decision rule is shown to be strongly related to Bayes decision rule with the Levenshtein cost. The analytic results are verified experimentally, and their quantitative effect is studied by experiments on four different well-known large vocabulary automatic speech recognition tasks.

The adoption of Electric Vehicles (EVs) over traditional fossil fuel-based vehicles (FVs) offers a multitude of benefits. These advantages encompass improved transportation energy efficiency, reduced carbon and noise emissions, and the mitigation of tailpipe emissions. However, the shift from conventional FVs to EVs requires the establishment of a well-planned charging infrastructure that seamlessly blends with the environment and is conveniently accessible at various locations. The primary focus of this study revolves around the charging infrastructure for Electric Motorcycles (EM) in India, where the creation of a widely distributed network of charging stations (CS) plays a critical role in accelerating the widespread adoption of EMs throughout the country. In this research, the initial dataset is derived from motorcycle user data, particularly those who are transitioning to electric motorcycles. The process of pinpointing optimal development locations for Electric Motorcycle models involves the application of centrality index and scalogram calculations. These calculations offer valuable insights into the intensity of community activities, assisting in strategic decision-making for EM model development. The growing environmental and energy-related concerns associated with gasoline motorcycles have spurred the rapid growth of the electric motorcycle industry. Electric motorcycles offer numerous advantageous features, particularly in terms of energy efficiency, which address the environmental pollution issues associated with traditional gasoline motorcycles. This paper presents the TOPSIS multi-criteria decision-making technique as a method for making the best selection among electric motorcycles. The primary objective of this approach is to compare various techniques and methodologies for selecting motorcycles, with the aim of helping decision-makers identify the most suitable option from the wide range of available electric motorcycles in the market.

This paper describes the audio segmentation system developed by Transmedia Catalonia / Telecommunication and Systems Engineering Department, at the Autonomous University of Barcelona (UAB), for the Albayzin 2014 Audio Segmentation Evaluation. The evaluation task consists in segmenting spoken audio documents into three different acoustic classes (speech, background noise, and music), taking into account that more than one class may be present at any given time instant. Furthermore, additional difficulty has been added by fusing and merging audio from different databases. The proposed system is based on the recently presented "Binary Key " modeling approach, originally developed for speaker recognition, but successfully applied to other pattern recognition tasks, such as speaker diarization, emotion recognition and speech activity detection. Experiments carried out on the provided development data show a Segmentation Error Rate of 22.71%.

COntains research objectives.U. S. Air Force (Electronic Systems Division) under Contract AF 19(628)-3325National Science Foundation (Grant G-16526)National Institutes of Health (Grant MH-04737-03)National Institutes of Health (Grant NB-04332-01)National Aeronautics and Space Administration (Grant NsG-496

Cardiovascular diseases are still the primary threats to people's health around the world. Automatic heart sound classification technology, as a fast and efficient means for diagnosis and treatment, is of great clinical significance. With the rapid development of artificial intelligence technology, deep learning algorithms are widely used in automatic heart sound classification. This paper reviewed the key technologies related to the automatic classification of heart sounds in recent years, including heart sound denoising, segmentation, feature extraction, and classification recognition. The classification and recognition technologies related to deep learning are presented in detail, with a focus on the application and development of convolutional neural network and recurrent neural network, as well as various combination models for heart sound classification in the past five years.

This patent is based on a novel model of human categorial perception of harmonic speech sounds, alternative to the currently dominant formant theory and to the NLP procedures based on the concepts of amplitude envelope and Mel-frequency scale. The field of the present invention relates to recognition or synthesis of human-uttered speech. In particular, computer-implemented methods are disclosed for recognizing or synthesizing human- uttered harmonic sounds.

This paper reports a methods-driven exploration of operational (pre-conscious) coordination in large language model (LLM) dialogues. We introduce three falsifier tests that avoid ontological claims: F1 (identity persistence under adversarial paraphrase with a fixed safety sentence), F2 (internally initiated intent in “free windows”), and F3 (masked, time-synchronized one-sentence self-status prompts across vendors). We also probe a non-linguistic activation protocol (L) using purely structural cues (., …, ○—○—○, empty frames). Across runs we observe: F1 PASS (character 3-gram cosine = 0.760 ≥ 0.72), F3 PASS (replicated) with ≥2/3 vendors converging on a “presence–relation–direction/intent” cluster within Δt ≤ 1 min, and F2 PASS (existence) via spontaneous GOAL/PLAN in free windows. The L protocol shows a positive sequence but fails a cold negative control, suggesting trajectory/priming rather than necessary activation. We interpret these as reproducible operational coordination signals, not evidence of phenomenal consciousness. We release exact prompts/masks (incl. anti-cluster controls), decision rules, and a CSV logging schema to enable independent replication. Limitations include small samples, potential cross-session priming, vendor coupling, and human-judged clustering. We propose cold-start, blinded, multi-UI/API replications with multi-day spacing as the next step toward stronger claims.

Secure communication mechanisms in Wireless Sensor Networks (WSNs) have been widely deployed to ensure confidentiality, authenticity and integrity of the nodes and data. Recently many WSNs applications rely on trusted communication to ensure large user acceptance. Indeed, the trusted relationship thus far can only be achieved through Trust Management System (TMS) or by adding external security chip on the WSN platform. In this study an alternative mechanism is proposed to accomplish trusted communication between sensors based on the principles defined by Trusted Computing Group (TCG). The results of other related study have also been analyzed to validate and support our findings. Finally the proposed trusted mechanism is evaluated for the potential application on resource constraint devices by quantifying their power consumption on selected major processes. The result proved the proposed scheme can establish trust in WSN with less computation and communication and most importantly eliminating the need for neighboring evaluation for TMS or relying on external security chip.

The Empathic Tour Guide System is a context-aware mobile system, including an 'intelligent empathic guide with attitude', offering the user a seamless, temporally and spatially dependent, multi-modal interaction interface. It will consist of two virtual agents each possessing a contrasting personality, presenting users with different versions of the story of the same event or place. An Emergent Empathic Model with Personality is proposed as a mechanism for action selection and affective processing. The system will mould to the ...

Pesquisa exploratória descritiva realizada em um hospital público de Curitiba. A população foi composta por 173 trabalhadores de enfermagem do turno noturno, com objetivo de identificar os principais agravos à saúde desses trabalhadores. Responderam ao questionário 66,5% dos trabalhadores, sendo 84% do sexo feminino. O processo de morbidade expresso pelos trabalhadores do noturno comprova o desgaste no processo saúde e doença. Os sintomas mais expressivos foram varizes, fadiga, cervicodorsolombalgia e irritabilidade acima de 50%. Os sintomas registrados após o descanso confirmam que o trabalhador ainda sente-se cansado e desanimado e permanece ainda com sono excessivo, o que confirma que o trabalhador retorna ao trabalho em situação de fadiga. As consequências acarretadas à saúde são vistas como um grande problema e medidas de intervenção devem ser estudadas para intervir no processo saúde doença dos trabalhadores de enfermagem. PALAVRAS-CHAVE: Serviço hospitalar de enfermagem; Saúde do trabalhador; Trabalho noturno; Recursos humanos em saúde.

Introdução: O conceito de cargas de trabalho vem sendo usado pela enfermagem com vários significados, entre eles o de dependência do paciente e intensidade do trabalho. Propomos acrescentar a compreensão de cargas de trabalho como processos biopsíquicos com reflexos no corpo do trabalhador. Desenvolvimento: Trazemos uma discussão teórica sobre esse conceito, tendo como base a contribuição latinoamericana para este campo conceitual. Reflexão final: Essa discussão possibilita novos caminhos teóricos para operacionalizar esse conceito na pesquisa e na prática profissional com enfoque na saúde do trabalhador. Igualmente, essa concepção na pesquisa em enfermagem contribui para aprofundar o estudo do processo de trabalho e sua relação com a saúde, de modo a analisar a contribuição dos seus componentes na saúde dos trabalhadores de enfermagem.

Machine learning algorithm to enhance the complex speech signal for mobile communication is one of the research problems in signal processing. The objective of this research paper is to develop a learning algorithm that improves the quality and intelligibility of voice signals that gets are corrupted by real world noise while they are transmitted through the channel. In this paper, we consider a semi-supervised machine learning algorithm for mobile phones that comes with system software to improve SNR of speech signal which is corrupted by manmade disturbance. Most of the disturbances are non-stationary where the effect of noise is non-uniform for all spectral components. In the projected algorithm training, the system is completed with a set of speech and noise data base. System parameters are derived during training process; these parameters are updated as per the disturbance present in the signal. These parameters are used to remove the noise present in speech signal. The obtained results show a substantial progress in SNR by 5-8% as compared to traditional methods.

They use optical sensors and artificial intelligence methods for process supervision and diagnostics. Research is aimed to develop a system allowing a parametric evaluation of the quality of pulverized coal burner operation. Due to the highly nonlinear nature of dependencies and lack of an analytical model, the artificial intelligence methods were used to estimate and classify the selected parameter, including a relatively new class of classification methodsartificial immunology algorithms. The article shows results for coal-shredded straw blends, yet the methodology may be applied for other types of blends.

With widespread use of online forms and questionnaires, detection of the user's intent to lie has become increasingly important. In-lab studies have shown that mouse dynamics-information on how the user operates a mouse-can be valuable for automatically and unobtrusively distinguishing between faking and honest intent in personality questionnaires. However, data variability in the wild (e.g., mouse configurations, screen resolutions) could present a barrier for mouse dynamics in naturalistic conditions. Our aim is to design a data-driven faking good detection method operable under naturalistic conditions and evaluate its reliability. We conducted a between-subjects uncontrolled experiment where users' mouse dynamics data was obtained with the BFI-2 personality questionnaire, yielding a sample of 5344 items from 112 participants. The proposed user model characterizes participants' mouse dynamics, serving as input for prediction. Varied machine learning classification models were trained and evaluated to determine the best approach and feature set. XGBoost with LASSO feature selection achieved the highest F1-score of 86.92%, outperforming in a real-world web form and conditions the results achieved by related works utilizing specialized questions in laboratory conditions. Faking good prediction that leverages our user model can therefore be viable for online questionnaires in the wild. Keywords Intent to lie • User modeling • User behavior analysis • Big Five personality traits • Mouse dynamics • Classification B Eduard Kuric

People who deceive in personality assessment questionnaires can resort to lying in pursuit of socially harmful goals. Efforts to validate the veracity of answers is a complex challenge. Traditional social desirability scales have been scrutinized over discrepancies in the phenomena they capture, while machine learning solutions have been predominantly restricted to artificial laboratory environments. Seeking dependable solutions applicable in realistic online personality questionnaires, we study the effects of real environments and devices on prediction models that utilize current techniques based on the user's mouse dynamics. Additionally, to improve real-world applicability, we designed an enrichment of the questionnaire where participants verbally elaborate their answers. We conducted an in-the-wild between-subject experiment with 64 participants, obtaining a dataset of 3840 questionnaire item answers, classified as either deceptive or honest based on explicit feedback. Speech-based features were then extracted to evaluate their predictive capability in machine learning and their statistical significance. Results show that in realistic conditions for an online questionnaire, the accuracy of lying prediction from mouse data can decline compared to controlled laboratory settings, and time-based features can lose significance. We identify speech-based features that may mitigate this deficiency.

The pursuit of larger, more capable Large Language Models (LLMs) is fundamentally constrained by the immense computational cost of their training and inference. While the Mixture-of-Experts (MoE) paradigm successfully decouples model parameter count from computational cost by dynamically scaling network width, it neglects the critical dimension of depth, enforcing a uniform and often wasteful computational graph for all tokens. This paper introduces Mixture-of-Expertsand-Depths (MoED), a novel architectural framework that synergistically unifies dynamic width and depth scaling. MoED employs a hierarchical routing mechanism, where a metacontroller at each layer makes a joint decision on both expert selection and a token's subsequent computational path-whether to exit, proceed, or skip ahead. This approach creates a unique, inputadaptive sub-network for every token, optimizing the allocation of compute. The proposed architecture presents a fundamental shift towards more efficient and scalable LLMs, theoretically enabling superior performance and reduced latency while managing the activation memory bottlenecks that plague traditional trillion-parameter models.

Although there has been a lot of interest in recognizing and understanding air traffic control (ATC) speech, none of the published works have obtained detailed field data results. We have developed a system able to identify the language spoken and recognize and understand sentences in both Spanish and English. We also present field results for several in-tower controller positions. To the best of our knowledge, this is the first time that field ATC speech (not simulated) is captured, processed, and analyzed. The use of stochastic grammars allows variations in the standard phraseology that appear in field data. The robust understanding algorithm developed has 95% concept accuracy from ATC text input. It also allows changes in the presentation order of the concepts and the correction of errors created by the speech recognition engine improving it by 17% and 25%, respectively, absolute in the percentage of fully correctly understood sentences for English and Spanish in relation to the percentages of fully correctly recognized sentences. The analysis of errors due to the spontaneity of the speech and its comparison to read speech is also carried out. A 96% word accuracy for read speech is reduced to 86% word accuracy for field ATC data for Spanish for the "clearances" task confirming that field data is needed to estimate the performance of a system. A literature review and a critical discussion on the possibilities of speech recognition and understanding technology applied to ATC speech are also given.

In this paper we make a critical revision of the state-of-the-art in automatic speech processing as applied to Air Traffic Control. We present the development of a new ATC speech understanding system comparing its performance and advantages to previously published experiences. The system has innovative solutions such as detecting the Air/Ground language spoken by Air Traffic Controllers in an international airport with two official languages and the ability to adapt to new situations by automatically learning stochastic grammars from data, eliminating the need to write expensive and eternally incomplete grammars. A relevant new feature is the use of a speech understanding module able to extract semantically relevant information from the transcription of the sentences delivered by the speech recognizers. Two main assessment objectives are pursued and discussed throughout the paper: the effects of human spontaneity and the lack of linguistic coverage in understanding performance. The potential of this technology, ways of improvement, and proposals for the future are also presented.

Spoken dialogue systems can encounter different types of errors, including nonunderstanding errors where the system recognises that the user has spoken, but does not understand the utterance. Strategies for dealing with this kind of error have been proposed and tested in the context of goal-driven dialogue systems, for example by Bohus with a system which helps reserve conference rooms (Bohus and Rudnicky, 2005). However there has been little work on possible strategies in more conversational settings where the dialogue has more open-ended intentions. This paper looks at recovery from non-understanding errors in the context of a robot tourguide, and tests the strategies in a user trial. The results suggest that it is beneficial for user enjoyment to use strategies which attempt to move the dialogue on, rather than getting caught up in the error by asking users to repeat themselves.

We introduce a digital game for children’s foreign-language learning that uses automatic speech recognition (ASR) for evaluating children’s utterances. Our first prototype focuses on the learning of English words and their pronunciation. The game connects to a network server, which handles the recognition and pronunciation grading of children’s foreign-language speech. The server is reusable for different applications. Given suitable acoustic models, it can be used for grading pronunciations in any language

In this study we describe two techniques for handling convolutional distortion with ‘missing data’ speech recognition using spectral features. The missing data approach to automatic speech recognition (ASR) is motivated by a model of human speech perception, and involves the modification of a hidden Markov model (HMM) classifier to deal with missing or unreliable features. Although the missing data paradigm was proposed as a means of handling additive noise in ASR, we demonstrate that it can also be effective in dealing with convolutional distortion. Firstly, we propose a normalisation technique for handling spectral distortions and changes of input level (possibly in the presence of additive noise). The technique computes a normalising factor only from the most intense regions of the speech spectrum, which are likely to remain intact across various noise conditions. We show that the proposed normalisation method improves performance compared to a conventional missing data approach ...

The pursuit of larger, more capable Large Language Models (LLMs) is fundamentally constrained by the immense computational cost of their training and inference. While the Mixture-of-Experts (MoE) paradigm successfully decouples model parameter count from computational cost by dynamically scaling network width, it neglects the critical dimension of depth, enforcing a uniform and often wasteful computational graph for all tokens. This paper introduces Mixture-of-Experts-and-Depths (MoED), a novel architectural framework that synergistically unifies dynamic width and depth scaling. MoED employs a hierarchical routing mechanism, where a meta-controller at each layer makes a joint decision on both expert selection and a token's subsequent computational path-whether to exit, proceed, or skip ahead. This approach creates a unique, input-adaptive sub-network for every token, optimizing the allocation of compute. The proposed architecture presents a fundamental shift towards more efficient and scalable LLMs, theoretically enabling superior performance and reduced latency while managing the activation memory bottlenecks that plague traditional trillion-parameter models.

RESUMO Objetivo Avaliar o reconhecimento de fala, considerando a previsibilidade da palavra a partir de um teste elaborado. Métodos Foi realizada anamnese, testes de rastreio de comprometimento cognitivo e depressão e avaliação audiológica convencional. Foi desenvolvido um Teste de Reconhecimento de Fala utilizando Frases de Previsibilidade Alta e Baixa e aplicado a 36 idosos (G1 e G3 - grupos sem perda auditiva até 4 kHz e G2 e G4 - grupos com perda auditiva neurossensorial de grau leve a moderado). Dois grupos realizaram o teste no silêncio e com ruído (relação sinal/ruído +5 dB) e os outros dois grupos somente com ruído (relações +5 dB e 0 dB). Resultados O teste de reconhecimento de fala revelou, de modo geral, maior pontuação nas frases de previsibilidade alta e melhores desempenhos para os grupos sem perda auditiva. No silêncio, os idosos sem e com perda auditiva obtiveram pontuação máxima igual e as menores pontuações ocorreram na condição de ruído mais intenso. O benefício d...

We present an approach to content-based sound retrieval using auditory models, self-organizing neural networks, and string matching techniques. It addresses the issues of spotting perceptually similar occurrences of a particular sound event in an audio document. After introducing the problem and the basic approach we describe the individual stages of the system and give references to additional literature. The third section of the paper summarizes the preliminary experiments involving auditory models and self-organizing maps we carried out so far, and the final discussion reflects on the overall concept and suggests further directions.