Speech Research Articles

Institutionalizing Speech Communication at the University of the Philippines: Epistemic and Embodied Practices of a Disciplinary Formation

Institutionalizing Speech Communication at the University of the Philippines: Epistemic and Embodied Practices of a Disciplinary Formation

Липкая инфляция, или языковая игра на материале словацких и российских экономических текстов

In modern linguistics, the understanding of language game refers to the field of speech communication, and the language games themselves are seen as a decoration of speech. Language gaming is multifarious. In this article we will focus on some forms of this phenomenon, manifested in the variation of words, their forms and meanings in Russian and Slovak economic media texts. In modern economic media texts, textual “expressivization” is observed, which is realized primarily with the help of the language games. Its evaluative function comes to the forefront – evaluativity is an integral category of media language. In our article we will focus on examples of language games in Slovak and Russian economic texts, which, in our opinion, contain this evaluation. Language gaming is a misuse of words or letters, a breach of logical correspondence, but, very importantly, it is an incorrectness that is realized by the author and deliberately allowed. Only a deliberate incorrectness will cause not annoyance and bewilderment, but a desire to support the game and try to uncover the author’s underlying intention.

Institutionalizing Speech Communication at the University of the Philippines: Epistemic and Embodied Practices of a Disciplinary Formation

Institutionalizing Speech Communication at the University of the Philippines: Epistemic and Embodied Practices of a Disciplinary Formation

Mandarin Recognition Based on Self-Attention Mechanism with Deep Convolutional Neural Network (DCNN)-Gated Recurrent Unit (GRU)

Speech recognition technology is an important branch in the field of artificial intelligence, aiming to transform human speech into computer-readable text information. However, speech recognition technology still faces many challenges, such as noise interference, and accent and speech rate differences. An aim of this paper is to explore a deep learning-based speech recognition method to improve the accuracy and robustness of speech recognition. Firstly, this paper introduces the basic principles of speech recognition and existing mainstream technologies, and then focuses on the deep learning-based speech recognition method. Through comparative experiments, it is found that the self-attention mechanism performs best in speech recognition tasks. In order to further improve speech recognition performance, this paper proposes a deep learning model based on the self-attention mechanism with DCNN-GRU. The model realizes the dynamic attention to an input speech by introducing the self-attention mechanism in a neural network model instead of an RNN and with a deep convolutional neural network, which improves the robustness and recognition accuracy of this model. This experiment uses 170 h of Chinese dataset AISHELL-1. Compared with the deep convolutional neural network, the deep learning model based on the self-attention mechanism with DCNN-GRU accomplishes a reduction of at least 6% in CER. Compared with a bidirectional gated recurrent neural network, the deep learning model based on the self-attention mechanism with DCNN-GRU accomplishes a reduction of 0.7% in CER. And finally, this experiment is performed on a test set analyzed the influencing factors affecting the CER. The experimental results show that this model exhibits good performance in various noise environments and accent conditions.

„Ненормативные” средства выражения адресованности в школьной практике, или игра с нормой

The article consists of several parts: an introduction, theoretical and practical parts, and the conclusion. This work touches upon several directions in linguistics. First of all, attention is paid to the concept of addressing, namely the terms addressee (Listener) and addresser (Speaker) and their interrelation at the moment of communication. The author provides a definition of human speech behavior, which at the present stage is usually considered in pragmalinguistics. The article focuses on the factor of the addresser in pedagogical discourse (the speech of a Russian language teacher at a secondary school in Slovakia is analyzed) from the point of view of observing the normativity of speech. In this regard, special attention is paid to the culture of speech and the concept of literary language. The work shows the varieties of norms of literary language, but the emphasis is on stylistic norms. The author focuses on the search for violations of stylistic norms in а scientific style. The article provides the author’s classification of selected language violations and characterizes the speech behavior of the Speaker.

Do external founder CEOs place strategic emphasis on innovation? An upper echelons perspective

AbstractResearch compares the influence of two types of chief executive officers (CEOs) on a firm's innovation strategy: founder CEOs, who founded the firms they lead, and professional CEOs, who worked in corporations at different levels before being selected as CEOs. We introduce a third type of CEO, external founders, who founded other ventures before being appointed as CEOs. Drawing on upper echelons theory, we propose that the unique combination of external founder CEOs' values and knowledge predisposes them to place less strategic emphasis on innovation than founder CEOs do but more than professional CEOs do. Heterogeneity among external founder CEOs may be due to nuances related to their exposure to professional investors in and successful exits from the ventures they founded. We assess our hypotheses empirically using the innovation‐related speech in investor communications and patent applications of 1637 CEOs of 789 S&P 500 firms from 2000 to 2019. We find that external founder CEOs emphasize innovation less than founder CEOs do, but contrary to our hypotheses, also less than professional CEOs do. However, certain nuances in external founder CEOs' founding experience bring their strategic emphasis on innovation close to that of founder CEOs. We extend upper echelons research by providing a fine‐grained view of the role of (founding) experience and its qualitative nuances in shaping CEOs' strategic decision‐making.

The Contribution of Artificial Intelligence Technologies in Refining the Arabic Language Skills of Its Native Speakers: Achieved and Possible

This study aims to shed light on the applications of Artificial Intelligence (AI) in improving Arabic language skills for speakers. It involves digitizing the language, developing computer programming, and adapting it to Arabic language learners to enhance the learning and teaching process. This study emphasizes the significance of implementing this project, as it can open up new horizons in educational curricula and teaching techniques. Moreover, it can contribute to spreading the use of the Arabic language and facilitating its acquisition, exchange, and dissemination globally, leading to positive outcomes on all levels. The current study also presents the most crucial benefits resulting from the digitization of the Arabic language, including supporting the self-learning strategy that has dominated most traditional methods. It also reinforces the status of the Arabic language in cognitive and scientific systems, blogs, and speech communication. One of the benefits is the easy acquisition and exchange of Arabic language skills, especially with the availability of various smartphone applications that have become indispensable in this era. Furthermore, the study addresses the significant challenges and difficulties facing this project and suggests appropriate solutions to overcome them. To what extent has the Arabic language benefited from this new digital transformation based on AI in the present era? And to what extent can the Arabic language rely on AI in the future so that its speakers can enhance their linguistic skills and proficiency in writing and speaking?

Attenuated processing of vowels in the left temporal cortex predicts speech-in-noise perception deficit in children with autism

BackgroundDifficulties with speech-in-noise perception in autism spectrum disorders (ASD) may be associated with impaired analysis of speech sounds, such as vowels, which represent the fundamental phoneme constituents of human speech. Vowels elicit early (< 100 ms) sustained processing negativity (SPN) in the auditory cortex that reflects the detection of an acoustic pattern based on the presence of formant structure and/or periodic envelope information (f0) and its transformation into an auditory “object”.MethodsWe used magnetoencephalography (MEG) and individual brain models to investigate whether SPN is altered in children with ASD and whether this deficit is associated with impairment in their ability to perceive speech in the background of noise. MEG was recorded while boys with ASD and typically developing boys passively listened to sounds that differed in the presence/absence of f0 periodicity and formant structure. Word-in-noise perception was assessed in the separate psychoacoustic experiment using stationary and amplitude modulated noise with varying signal-to-noise ratio.ResultsSPN was present in both groups with similarly early onset. In children with ASD, SPN associated with processing formant structure was reduced predominantly in the cortical areas lateral to and medial to the primary auditory cortex, starting at ~ 150—200 ms after the stimulus onset. In the left hemisphere, this deficit correlated with impaired ability of children with ASD to recognize words in amplitude-modulated noise, but not in stationary noise.ConclusionsThese results suggest that perceptual grouping of vowel formants into phonemes is impaired in children with ASD and that, in the left hemisphere, this deficit contributes to their difficulties with speech perception in fluctuating background noise.

A corollary discharge circuit in human speech

When we vocalize, our brain distinguishes self-generated sounds from external ones. A corollary discharge signal supports this function in animals; however, in humans, its exact origin and temporal dynamics remain unknown. We report electrocorticographic recordings in neurosurgical patients and a connectivity analysis framework based on Granger causality that reveals major neural communications. We find a reproducible source for corollary discharge across multiple speech production paradigms localized to the ventral speech motor cortex before speech articulation. The uncovered discharge predicts the degree of auditory cortex suppression during speech, its well-documented consequence. These results reveal the human corollary discharge source and timing with far-reaching implication for speech motor-control as well as auditory hallucinations in human psychosis.

Speech Emotional Recognition of Telephone Conversation by Using Deep Learning

In this research, we have compare two clustering algorithms, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) to support the SER model for voice communication. We proposed an approach to speech emotion recognition in telephone conversations using a combination of Mel-frequency cepstral coefficients (MFCC) for audio feature extraction. Only the zeroth coefficient (energy) of MFCC will be used, as energy can provide a good representation for sound. The extracted results are then classified using CNN and RNN. The CNN algorithm consistently achieved a higher accuracy than RNN with the values of 0.93 at epoch=50, 0.93 at epoch=100, 0.90 at epoch=150, and 0.93 at epoch=200. But the RNN algorithm has a faster training times than CNN. For optimizer test, Adam optimizer performs well for both models with respective accuracy values of 0.93 and 0.94, outperforming other optimizers for accuracy.

Vocal Speed and Processing of Persuasive Messages: Curvilinear Processing Effects

AbstractMost work on indicators of vocal confidence (and social influence work more broadly) examines linear relationships between variables. However, in some domains curvilinear (i.e., accelerating or decelerating) relationships may provide greater clarity in understanding human speech patterns. We review mixed past work on vocal speed as a case study, wherein faster vocal speed has been shown both to bolster and inhibit persuasion (e.g., by impairing processing). Across six total studies (Ntotal = 3,958), we show that faster speed initially increases perceived source confidence and message processing but eventually the increase attenuates or reverses. Correspondingly, vocal speed has a decelerating relationship to participants’ processing of persuasive messages, as revealed by two main processes: argument quality effects on attitudes, and the correspondence between thought valence and attitudes. The present work highlights the potential value of high-powered examinations of curvilinear relationships in non-verbal phenomenon for which speed is likely to play a role.

Chaos-based Cryptography for Voice Communication Security

Chaos-based cryptography represents a groundbreaking approach to securing voice communications, leveraging chaotic systems to integrate unpredictability and randomness into encryption keys. This innovative methodology protects against eavesdropping, man-in-the-middle (MitM) attacks, and spyware, safeguarding sensitive conversations. By utilizing chaos theory to encrypt voice signals, Chaos- based cryptography strictly limits access to authorized parties, maintaining confidentiality and data integrity. Notably, this chaos-based encryption framework delivers exceptional voice transmission speed and reliability, making it ideal for critical applications demanding high communication velocity and precision. Chaos- based cryptography also adheres to regulatory standards for secure communication, guaranteeing compliance across various industries. Consequently, this cutting-edge technology is highly sought after for enhancing voice communication system security. The study presents an overview of the chaos- based cryptography for voice communication security.

Methods of Voice Restoration following Laryngectomy

In Germany, approximately 950-1200 laryngectomies are performed annually due to extensive hypopharyngeal and laryngeal carcinomas. The removal of the larynx results in significant life changing functional losses, particularly in voice production and communication, resulting in psychological issues and social withdrawal. Rehabilitation requires a multidisciplinary team focusing on restoring communication skills. Methods include pseudowhispering, esophageal speech, external mechanical and electromechanical aids, and tracheoesophageal speech prostheses. Pseudowhispering uses air from the mouth and pharynx, while esophageal speech requires controlled release of esophageal air to produce sound. External aids include devices like the Electrolarynx. Tracheoesophageal speech prostheses, the current gold standard, offer superior phonation time and clarity but require frequent visits to the healthcare system to address complications due to the limited lifespan of the prosthesis. Despite high success rates with transoesophageal prostheses, some patients may face significant challenges. In conclusion, voice rehabilitation post-laryngectomy is crucial for social reintegration. Comprehensive patient education on the benefits and challenges of different methods is essential for making an informed decision tailored to their lifestyle.

2-D deployment of aerial base stations: A simulation model to provide voice communication

Unmanned aerial vehicles (UAVs) offer a potential alternative for providing voice services in areas where communication is disrupted due to natural disasters. These UAVs can be configured as aerial base stations (ABSs), enabling the deployment of a temporary communications network. However, communication networks based on ABSs pose several significant challenges. One of these challenges involves addressing interruptions or limitations in network coverage caused by natural disasters. In such situations, there is a high likelihood that users within the affected area may be unable to communicate due to a lack of coverage. This is a complex problem because it depends on factors, such as the mobile user locations, the characteristics of the air-to-ground channel, and geographical details of the area. In this work, we propose an optimization model to determine the placement of a set of ABSs within a limited disaster area that maximizes the probability of successful voice services (PSVSs). This optimization model integrates a network evaluation model that analyzes the wireless environment at a specific time. The network evaluation model utilizes two-ray and Rayleigh channel models, enabling the simulation of a worst-case scenario for wireless communication systems. We evaluate the proposed optimization model using the (1+1)-evolution strategy with a one-fifth success rule. We explore various parameter configurations to understand their impact on algorithm performance. This analysis helps identify the configuration of the optimization model that yields the maximum PSVSs. Simulation results indicate that by appropriately configuring the evolution strategy algorithm and comparing random ABS locations with those determined ABS locations by the evolution strategy algorithm, the PSVS can be enhanced by an average of 60%.

Identificación del acento en hablantes de español mediante el análisis de atributos MFCC y aprendizaje supervisado

Speaker recognition has multiple real-life applications. The purpose of this study is to determine the feasibility of classifying samples of human speech, specifically Spanish speakers, based on their distinctive accents. In this work, Mel-Frequency Cepstral Coefficients (MFCC) combined with machine learning techniques were used to identify the nationality of Spanish-speaking individuals through voice recordings obtained from the Crowdsourcing Latin American Spanish for Low-Resource Text-to-Speech corpus. Data preprocessing was performed by extracting 50 MFCC from each recording, which formed the dataset for experimentation. Experiments were conducted with different subsets, and the best results were obtained with individuals from four Latin American countries, including both males and females. Neural networks were employed for the classification stage, achieving an accuracy of 99.84%.

Measuring Hearing Aid Satisfaction in Everyday Listening Situations: Retrospective and In Situ Assessments Complement Each Other.

Recently, we developed a hearing-related lifestyle questionnaire (HEARLI-Q), which asks respondents to rate their hearing aid (HA) satisfaction in 23 everyday listening situations. It is unknown how HA satisfaction on the retrospective HEARLI-Q scale compares with HA satisfaction measured on the same scale implemented in Ecological Momentary Assessment (EMA). To learn how retrospective (HEARLI-Q) and in situ (EMA) assessments can complement each other. An observational study. Twenty-one experienced HA users. The participants first filled out the HEARLI-Q questionnaire, followed by a 1-week EMA trial using their own HAs. HA satisfaction ratings were compared between the two questionnaires and the underlying drivers of discrepancies in HA satisfaction ratings were evaluated. HA satisfaction ratings were significantly higher in EMA for speech communication with one or several people. Hearing difficulty in these situations was rated higher in HEARLI-Q than in EMA, but occurrence of those difficult listening situations was also rated to be lower. When comparing only the situations that occur on daily or weekly basis, the two questionnaires had similar HA satisfaction ratings. Lower occurrence of difficult listening situations seems to be the key driver of discrepancies in HA satisfaction ratings between EMA and HEARLI-Q. The advantage of EMA is that it provides insight into an individual's day-to-day life and is not prone to memory bias. HEARLI-Q, on the other hand, can capture situations that occur infrequently or are inconvenient to report in the moment. Administering HEARLI-Q and EMA in combination could give a more holistic view of HA satisfaction.

Design and Development of an Acoustic-Based Recongnition System Using DNN

Automatic speech recognition is a process of using computers to convert voice signals produced by human speech into reasonable format i.e. text or command that conveys the same meaning as the speaker intended to do. Many researchers are working on various languages including English and other European languages like Spanish, German, and French etc. to develop an automated system for speech recognition (ASR). However, researchers on the development of ASR for the Urdu language have put very little effort. We have developed an Urdu speech recognition system using Deep Neural Network (DNN) on our developed corpus that contains some of the most frequently used words in Urdu like digits, season names, and month names. The accuracy rates of our ASR are very encouraging because 72% accuracy is achieved for 26 words and 92% accuracy is achieved separately for names of seasons.

Expression and regulation of SETBP1 in the song system of male zebra finches (Taeniopygia guttata) during singing

Rare de novo heterozygous loss-of-function SETBP1 variants lead to a neurodevelopmental disorder characterized by speech deficits, indicating a potential involvement of SETBP1 in human speech. However, the expression pattern of SETBP1 in brain regions associated with vocal learning remains poorly understood, along with the underlying molecular mechanisms linking it to vocal production. In this study, we examined SETBP1 expression in the brain of male zebra finches, a well-established model for studying vocal production learning. We demonstrated that zebra finch SETBP1 exhibits a greater number of exons and isoforms compared to its human counterpart. We characterized a SETBP1 antibody and showed that SETBP1 colocalized with FoxP1, FoxP2, and Parvalbumin in key song nuclei. Moreover, SETBP1 expression in neurons in Area X is significantly higher in zebra finches singing alone, than those singing courtship song to a female, or non-singers. Importantly, we found a distinctive neuronal protein expression of SETBP1 and FoxP2 in Area X only in zebra finches singing alone, but not in the other conditions. We demonstrated SETBP1´s regulatory role on FoxP2 promoter activity in vitro. Taken together, these findings provide compelling evidence for SETBP1 expression in brain regions to be crucial for vocal learning and its modulation by singing behavior.

Intelligent Data Management for Small Business Enhancement

In the industry 4.0 era, there exists a pressing need for intelligent data management solutions to enhance the operations of small businesses. This study introduces a pioneering methodology that harnesses the power of AI-driven analysis of internal voice communications, an often-overlooked source of valuable insights within the small business environment. The research centers on an advanced platform that utilizes the Regularized Bayesian Approach, meticulously tailored for the processing of unstructured and semi-structured data, with a specific focus on internal voice messages. This methodology enables the generation of in-depth insights into employees’ emotional, psychological, and motivational states. Furthermore, the integration of data with a psychometric system enables the production of comprehensive personality evaluations, providing digital portraits for every employee. These portraits offer valuable insights into employee well-being and motivations, particularly beneficial for small businesses with limited HR resources. The potential benefits for small businesses are multifaceted and research-driven, including enhanced employee safety, improved efficiency, advanced risk management, and streamlined HR processes. Additionally, this research underscores the growing relevance and potential of this approach in the Emotion AI market. Through the analysis of voice messages, entities, intent, and relationships between utterances can be discerned, offering a comprehensive view of employee sentiment, loyalty, and satisfaction. This study serves as the foundation for fostering a positive work environment, enhancing productivity, and providing a roadmap for mental health improvement and reduced attrition in small businesses. It contributes to the evolving field of intelligent data management and its applications in enhancing small business operations. Keywords: voice recognition, small business enhancement, emotion AI, artificial intelligence, Bayesian approach

INTELLIGENT SPEECH RECOGNITION USING FRACTAL AMENDED GRASSHOPPER OPTIMIZATION ALGORITHM WITH DEEP LEARNING APPROACH

Humans prefer to convey information through speech utilizing similar language. Speech detection is the capability to recognize the spoken words of the speaking person. Recent work demonstrates the increased attention among researcher workers in this field specially in Brain-Like computing applications and emphasizes the real-world usability of speech for speaker recognition across different applications. Automatic speech recognition (ASR) is the method of identifying human speech and converting it into text. This study has gained much popularity in recent times. It is a crucial area of research for human-to-machine interaction. Pioneer methods are concerned with manual feature extraction and classical algorithms including Hidden Markov Models (HMM), Gaussian Mixture Model (GMM), and the Dynamic Time Warping (DTW) model. In recent years, neural networks, namely convolutional neural networks (CNN), recurrent neural networks (RNN), and Transformers, have been utilized in the context of ASR and reached outstanding performance over the past few years. This study introduces Intelligent Speech Recognition using the Fractal Amended Grasshopper Optimization Algorithm with Deep Learning (ISR-AGODL) approach. The presented ISR-AGODL technique correctly identifies and recognizes speech signals. In the ISR-AGODL technique, the speech signals are transformed into spectrograms. Besides, the features are derived using the deep convolutional neural networks (DCNN) model. Followed by the Fractals AGO technique is utilized for the choosing of hyperparameters. Finally, the recognition of speech signals is achieved using the extreme gradient boosting (XGBoost) model. The simulation outcomes of the ISR-AGODL method can be validated using a benchmark dataset. The experimental results of the ISR-AGODL method portrayed a superior accuracy outcome of 96.34% over other models.