Search

Introduction: Many studies have claimed that speech perception in dyslexic children is impaired in comparison to normal children in a way that dyslexic children are weaker in the discrimination and identification of phonemes. In this research, we study speech perception, syntactic skills, and phonological processing in the normal and dyslexic Farsi-speaking children. Materials and Methods: We examined the speech perception in 30 normal and 18 dyslexic children, using speech identification task in 3 continua of Voice Onset Time (VOT) in [ba-pa], [da-ta], and [ga-ka]. We also studied 2 other skills to examine the processing of phonological information and syntactic skills. We used the nonword repetition test and passive sentence comprehension, respectively. To compare the data, the Mann-Whitney U test, t-test, and logistic regression were used. Results: Dyslexic children in speech identification task performed similarly to normal children in all 3 continua (P=0.81, 0.87, and 0.58); but, in the nonword repetition test, they were significantly lower than normal children (P=0.01), and in the passive sentence comprehension task, there was no significant difference between normal and dyslexic children (P=0.12). Conclusion: The results of various studies for speech perception and linguistic abilities in dyslexic children are contradictory. They are mostly related to the tasks that have been used. Therefore, more studies in the future are suggested. [ABSTRACT FROM AUTHOR]

This paper surveys human-inspired speech technologies developed for European Portuguese and the computational models they integrate and made them possible. In this regard, it covers systems for synthesis and recognition as well as information on the methods adopted for the speech production studies that were performed, in parallel, to support them. And, on doing so, it can also contribute to provide an entry point for those who work in the field but are not familiar with these particular areas, including: context, history, and comprehensive references. As the great majority of work in these areas for European Portuguese was done by the first author's research group, this paper can also be seen as a review of more than 25 years of research at University of Aveiro in these topics. [ABSTRACT FROM AUTHOR]

Human–computer interaction, Artificial Intelligence, and the multilingualism of digital culture open up unexpected scenarios in the contemporary design of cultural spaces with the creation of artifacts in which the analogic and digital dimensions come together to enhance the experience. Sensors and devices track user movement in the real world and translate the inputs into commands through hand gestures, speech recognition, head movements, tangible interfaces, or a combination of these elements. Through theorical models, concepts and tools, the paper reports the evolution of "User Experience" applied to personalized enjoyment and use of cultural places. The functional-performance survey of tools and technologies for perception, narrative and augmented interaction revealed models that highlight the diversity and richness of tangible and intangible cultural heritage through new forms of interaction and knowledge transfer. In this direction, new technological tools make it possible to detect, track and evaluate the personalized user experience by processing or producing large amounts of data. Therefore, on one hand the paper explores the limitations dictated by data management and user privacy in using such systems and, on the other, it prefigures new scenarios for amplifying and personalizing the user experience. [ABSTRACT FROM AUTHOR]

With the rapid development of information technology in modern society, the application of multimedia integration platform is more and more extensive. Speech recognition has become an important subject in the process of multimedia visual interaction. The accuracy of speech recognition is dependent on a number of elements, two of which are the acoustic characteristics of speech and the speech recognition model. Speech data is complex and changeable. Most methods only extract a single type of feature of the signal to represent the speech signal. This single feature cannot express the hidden information. And, the excellent speech recognition model can also better learn the characteristic speech information to improve performance. This work proposes a new method for speech recognition in multimedia visual interaction. First of all, this work considers the problem that a single feature cannot fully represent complex speech information. This paper proposes three kinds of feature fusion structures to extract speech information from different angles. This extracts three different fusion features based on the low-level features and higher-level sparse representation. Secondly, this work relies on the strong learning ability of neural network and the weight distribution mechanism of attention model. In this paper, the fusion feature is combined with the bidirectional long and short memory network with attention. The extracted fusion features contain more speech information with strong discrimination. When the weight increases, it can further improve the influence of features on the predicted value and improve the performance. Finally, this paper has carried out systematic experiments on the proposed method, and the results verify the feasibility. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) is a condition that degenerates over time and impairs speech and pronunciation because brain cells have died. This research work aims to predict parkinson disease using the voice features extracted from speech signals recorded from PD individuals with dysphonic speech disorders by employing deep learning algorithms. PD is challenging to diagnose early on in the clinical presentation. To address the issue in machine learning methods, this paper proposes a neural network model by processing speech signals to classify PD using the University of California Irvine (UCI) machine learning repository dataset. Initially, a pre-loss reduction module is created by using pre-sampling to make the dataset balanced by reducing the dimensionality and maintaining the size of the space without influencing the learning process for data preparation. The relevant features are derived using a novel multi-agent salp swarm (MASS) algorithm, and a novel Parkinson classification neural network (PCNN) is proposed to classify Parkinson's patients with high accuracy employing these derived features. The result shows that the models that use MASS-PCNN produce higher classification accuracy of 99.1%, precision of 97.8%, recall of 94.7% and F1-score of 0.995 when paralleled to the existing models. As an outcome, the suggested model will perform superior to common convolutional neural networks. [ABSTRACT FROM AUTHOR]

The extraction of weak acoustic signals under strong background noise is of great significance in the applications of target identification and localization. In this paper, the pulse signal with high randomness is set as the weak signal sound source, random noise and sine sound are used as the background noise. Under the condition of a signal-to-noise ratio of −20 dB, combined with blind source separation and neural network methods, the collected observation signals are subjected to weak sound signal separation and recognition research. The optimization method of centralization and scaling processing is used to eliminate the unfavorable influence of the uncertainty of the separated signal amplitude caused by the blind source separation method on the pattern recognition. The recognition result is verified by the combination of "weak impulse acoustic signal" and "random noise signal," and the output vector (0.99 0.01 0.01) approaches (1 0 0), which is recognized as impulse acoustic signal. By combining blind source separation and neural network methods, the separation and identification of weak pulse signals under the condition of a signal-to-noise ratio of −20 dB can be achieved. [ABSTRACT FROM AUTHOR]

Prosodic aspects of cross-linguistic contact are under-researched, especially past contact that has subsequently ceased. In this paper, we investigate declarative and polar question tunes of contemporary Cretan Greek, a regional variety of Greek whose speakers were in contact with Venetian speakers during the four and half centuries of Venetian rule on the island, from 1204 to 1669. The F0 contours of the Cretan tunes and alignment of peaks and troughs of interest with the nuclear vowel are compared to the corresponding tunes in Venetian dialect and Venetian Italian and to those in Athenian (Standard) Greek, which are used as control. The data (1610 declarative utterances and 698 polar questions) were drawn from natural speech corpora based on pragmatic criteria: broad focus for declaratives, broad focus, and information-seeking interpretation for polar questions. The pitch contour shapes of the tunes are modeled using polynomial basis functions, and the F0 alignment points are determined analytically. The results show the robustness of contact effects almost three and a half centuries after regular contact ceased and indicate that the shapes of the F0 contours of Cretan and Venetian declarative and polar question tunes are similar. In addition, Cretan alignment patterns are similar to Venetian and significantly different from Athenian. Insights are gained from research into how long prosodic characteristics may persist in a recipient language—decades or even centuries after the cessation of contact. [ABSTRACT FROM AUTHOR]

Aims and objectives: This study examines how social information is utilized in processes of bilingual speech perception. Specifically, we investigate whether racialized expectations of native language background trigger language-specific processing strategies in early or simultaneous Spanish–English bilinguals. Methodology: We coupled a visually-primed phoneme categorization task with a social evaluation questionnaire to test whether Spanish–English bilingual adults living in the United States (n = 30) drew on racialized ideologies about what speakers of certain languages look like during speech perception. We predicted that, if participants drew on these ideologies during the phoneme categorization task, they would cue an expectation of what language was being spoken and, consequently, a shift in the identification boundary. Data and analysis: Mixed logistic regression was used to investigate the effects of photograph (the visual prime) and voice on how participants categorized the continua while paired, two-tailed t -tests were used to compare how participants socially evaluated the two speakers. Findings: Raciolinguistic evaluations appeared to influence bilingual speech perception, significantly affecting how the continua were categorized, but they did not work the same way for every voice. Originality: Prior work has posited that interactional context influences bilingual language control (e.g., Grosjean, 2001); the ideological nature of this context has, however, been understudied. This paper offers insight into how ideologies related to race may shape language perception and use in bilinguals. Implications: The findings provide evidence for the role of social information in bilingual speech perception, suggest that multiple cues (acoustic and social) integrate to determine the interactional context, and indicate that the influence of raciolinguistic ideologies is neither straightforward nor homogeneous but rather contingent on complex aspects of the perceived speaker. [ABSTRACT FROM AUTHOR]

Partial discharge (PD) is a crucial and intricate electrical occurrence observed in various types of electrical equipment. Identifying and characterizing PDs is essential for upholding the integrity and reliability of electrical assets. This paper proposes an ensemble methodology aiming to strike a balance between the model complexity and the predictive performance in PD pattern recognition. A simple convolutional neural network (SCNN) was constructed to efficiently decrease the model parameters (quantities). A quadratic support vector machine (QSVM) was established and ensembled with the SCNN model to effectively improve the PD recognition accuracy. The input for QSVM consisted of the circular local binary pattern (CLBP) extracted from the enhanced image. A testing prototype with three types of PD was constructed and 3D phase-resolved pulse sequence (PRPS) spectrograms were measured and recorded by ultra-high frequency (UHF) sensors. The proposed methodology was compared with three existing lightweight CNNs. The experiment results from the collected dataset emphasize the benefits of the proposed method, showcasing its advantages in high recognition accuracy and relatively few mode parameters, thereby rendering it more suitable for PD pattern recognition on resource-constrained devices. [ABSTRACT FROM AUTHOR]

Large vision-language models, such as Contrastive Vision-Language Pre-training (CLIP), pre-trained on large-scale image–text datasets, have demonstrated robust zero-shot transfer capabilities across various downstream tasks. To further enhance the few-shot recognition performance of CLIP, Tip-Adapter augments the CLIP model with an adapter that incorporates a key-value cache model constructed from the few-shot training set. This approach enables training-free adaptation and has shown significant improvements in few-shot recognition, especially with additional fine-tuning. However, the size of the adapter increases in proportion to the number of training samples, making it difficult to deploy in practical applications. In this paper, we propose a novel CLIP adaptation method, named Proto-Adapter, which employs a single-layer adapter of constant size regardless of the amount of training data and even outperforms Tip-Adapter. Proto-Adapter constructs the adapter's weights based on prototype representations for each class. By aggregating the features of the training samples, it successfully reduces the size of the adapter without compromising performance. Moreover, the performance of the model can be further enhanced by fine-tuning the adapter's weights using a distance margin penalty, which imposes additional inter-class discrepancy to the output logits. We posit that this training scheme allows us to obtain a model with a discriminative decision boundary even when trained with a limited amount of data. We demonstrate the effectiveness of the proposed method through extensive experiments of few-shot classification on diverse datasets. [ABSTRACT FROM AUTHOR]

Speech is produced by a nonlinear, dynamical Vocal Tract (VT) system, and is transmitted through multiple (air, bone and skin conduction) modes, as captured by the air, bone and throat microphones respectively. Speaker specific characteristics that capture this nonlinearity are rarely used as stand-alone features for speaker modeling, and at best have been used in tandem with well known linear spectral features to produce tangible results. This paper proposes Recurrent Plot (RP) embeddings as stand-alone, non-linear speaker-discriminating features. Two datasets, the continuous multimodal TIMIT speech corpus and the consonant-vowel unimodal syllable dataset, are used in this study for conducting closed-set speaker identification experiments. Experiments with unimodal speaker recognition systems show that RP embeddings capture the nonlinear dynamics of the VT system which are unique to every speaker, in all the modes of speech. The Air (A), Bone (B) and Throat (T) microphone systems, trained purely on RP embeddings perform with an accuracy of 95.81%, 98.18% and 99.74%, respectively. Experiments using the joint feature space of combined RP embeddings for bimodal (A–T, A–B, B–T) and trimodal (A–B–T) systems show that the best trimodal system (99.84% accuracy) performs on par with trimodal systems using spectrogram (99.45%) and MFCC (99.98%). The 98.84% performance of the B–T bimodal system shows the efficacy of a speaker recognition system based entirely on alternate (bone and throat) speech, in the absence of the standard (air) speech. The results underscore the significance of the RP embedding, as a nonlinear feature representation of the dynamical VT system that can act independently for speaker recognition. It is envisaged that speech recognition too will benefit from this nonlinear feature. [ABSTRACT FROM AUTHOR]

With the development of deep learning techniques, the field of direction of arrival (DOA) estimation has also made significant progress. However, the accuracy of DOA estimation using end-to-end neural networks (NNs) heavily relies on the classification step of the networks, which necessitates the use of large and representative datasets. Additionally, conventional speech presence probability (SPP) estimation methods based on the ideal ratio mask (IRM) may misclassify time-frequency (T-F) bins dominated by non-speech and noise, which hinders the accurate extraction of directional information. To improve the robustness of existing DOA estimation algorithms, this paper proposes a DOA estimation method with T-F bin selection. In terms of output, instead of using IRM-based SPP, our proposed approach focuses on the a posteriori SPP, a deliberate choice aimed at circumventing potential confusion. For input optimization, we construct features that encompass spatial, temporal, and directional information concurrently, and these are coupled with a frequency bin-wise recurrent neural network (RNN) model to attain precise multi-channel SPP estimation. Subsequently, these SPP estimates are utilized to extract local information for DOA estimation. Moreover, the cascaded structure ensures that the model has the ability to complete out-of-label tasks, effectively reducing the dataset requirements by training only a subset of direction information to achieve omnidirectional DOA estimation. Besides, this contributes to the algorithm's ability to eliminate its reliance on the step size, setting it apart from other end-to-end methods. Simulation results validate that the proposed method achieves higher accuracy and lower error compared to both NN-based end-to-end approaches and traditional full-band approaches under various conditions of reverberation and signal-to-noise ratio. [ABSTRACT FROM AUTHOR]

Background: People with aphasia (PWA) frequently experience difficulties in understanding longer written content such as paragraphs or books. Reading strategies are a promising approach to treat text‐level reading comprehension deficits in PWA. Nevertheless, empirical evidence for their efficacy remains rare. Aims: The primary objective of this study was to analyse the efficacy of a strategy‐based intervention on text‐level reading comprehension in PWA. Secondary objectives were to compare the effects of two strategy‐based intervention components and to explore potential moderator effects. Methods & Procedures: A protocol was published prior to data acquisition. In a repeated measures trial, 26 German participants with chronic, post‐stroke aphasia participated in a waiting period without aphasia treatment (control condition) followed by a strategy‐based intervention called 'Strategiebasierte Textverständnis‐Therapie bei Aphasie' (StraTexT, 14 face‐to‐face‐sessions, twice per week, 60 min each). Two strategy combinations, Intervention Micro targeting microstructure and Intervention Macro targeting macrostructure, were applied to newspaper and magazine articles. Participants were randomly allocated to two parallel groups that received these strategy combinations in interchanged sequences. Assessments were implemented before and after each period as well as 3 and 6 months after the intervention. The primary outcome measure was text‐level reading comprehension measured with the total score of a German version of the Test de Compréhension de Textes (TCT‐D). Secondary outcome measures addressed the self‐reported perception of reading abilities, reading activities and feelings about reading (German version of the Comprehensive Assessment of Reading in Aphasia CARA reading questionnaire) as well as selected cognitive functions. Outcomes & Results: The per‐protocol‐analysis included data from 22 participants. We found significant small improvements up to 6 months post‐intervention compared to pre‐intervention in the TCT‐D Total (d = 0.35–0.46) as well as medium to large improvements in the CARA questionnaire (d = 0.68–0.96). Up to 3 months after the intervention, treatment‐induced improvements in the TCT‐D Total were significantly larger than change without treatment during the control condition. There was no evidence of moderator effects. Furthermore, we found improvements in several cognitive functions. Conclusions & Implications: Reading strategies can lead to long‐term improvements in text‐level reading comprehension and in self‐reported reading abilities, feelings about reading and reading activities in aphasia. In regular clinical settings, it seems reasonable to implement both Intervention Micro and Intervention Macro. It remains important to investigate participant characteristics that contribute to treatment success. WHAT THIS PAPER ADDS: What is already known on the subject: Systematic reviews and multiple case studies suggest that reading strategies are a promising approach to treat text‐level reading comprehension in aphasia. The efficacy of reading strategies has been demonstrated for different populations. However, to date no group study has evaluated the efficacy of reading strategies on text‐level reading comprehension in people with aphasia. What this study adds: This study provides the first group‐level evidence about the efficacy of a systematic strategy‐based intervention in 22 people with post‐stroke chronic aphasia. During 14 treatment sessions, participants applied four reading strategies to newspaper and magazine articles within two intervention components called Intervention Micro and Intervention Macro (two strategies per intervention component). We found improvements in text‐level reading comprehension for at least 3 months post‐intervention, as well as effects on selected cognitive functions and self‐reported reading abilities, thoughts and feelings about reading and the ability to engage in reading activities. What are the clinical implications of this work?: The strategies and materials evaluated in this study could be used in clinical practice with people with aphasia. In order to replicate treatment effects in clinical practice, we suggest applying the strategy combination with the same protocol features (e.g., frequency, duration, homework, product orientation) as in this study, implementing Intervention Micro and Intervention Macro sequentially in either order. As treatment response was not equal in all individuals, it seems important to investigate individual features that contribute to treatment success. [ABSTRACT FROM AUTHOR]

Background: Children with phonological impairment present with pattern‐based errors in their speech production. While some children have difficulties with speech perception and/or the establishment of robust underlying phonological representations, the nature of phonological impairment in children is still not well understood. Given that phonological and lexical development are closely linked, one way to better understand the nature of the problem in phonological impairment is to examine word learning abilities in children. Aims: To examine word learning and its relationship with speech perception, speech production and vocabulary knowledge in children aged 4–5 years. There were two variables of interest: speech production abilities ranging from phonological impairment to typical speech; and vocabulary abilities ranging from typical to above average ('lexically precocious'). Methods & Procedures: Participants were 49 Australian‐English‐speaking children aged 48–69 months. Children were each taught four novel non‐words (out of a selection of eight) through stories, and word learning was assessed at 1 week post‐initial exposure. Word learning was assessed using two measures: confrontation naming and story retell naming. Data were analysed by group using independent‐samples t‐tests and Mann–Whitney U‐tests, and continuously using multiple linear regression. Outcomes & Results: There was no significant difference in word learning ability of children with and without phonological impairment, but regardless of speech group, children with above average vocabulary had significantly better word learning abilities than children with average vocabulary. In multiple linear regression, vocabulary was the only significant predictor of variance in word learning ability. Conclusions & Implications: Children with phonological impairment can be lexically precocious and learn new words like their peers without phonological impairment. Contrary to expectations, vocabulary knowledge rather than expressive phonological ability explained variance in measures of word learning. These findings question an assumption that children with phonological impairment have underspecified phonological representations. They also highlight the heterogeneity among children with phonological impairment and the need to better understand the nature of their difficulty learning the phonological system of the ambient language. WHAT THIS PAPER ADDS: What is already known on the subject: There is limited research examining the word learning abilities of children with phonological impairment. Most previous research focuses on word properties such as phonotactic probability and neighbourhood density. Within the existing literature there are different reports and conclusions regarding the word learning abilities of children with phonological impairment and whether their word learning differs from that of children with typically developing speech. What this study adds to existing knowledge: This study found that vocabulary was the strongest predictor of word learning across children with and without phonological impairment. There was no significant difference in word learning ability between children with and without phonological impairment. However, children with lexically precocious vocabulary abilities were significantly better at word learning than children with average vocabulary abilities. What are the potential or actual clinical implications of this work?: Findings from this study support the importance of assessing and considering measures of word learning—including vocabulary—when working with children with phonological impairment. This study indicates that it is possible to use stories coupled with measures of confrontation naming and story retell to gain deeper insight into children's word learning abilities. [ABSTRACT FROM AUTHOR]

While deep neural networks have shown impressive results in automatic speaker recognition and related tasks, it is dissatisfactory how little is understood about what exactly is responsible for these results. Part of the success has been attributed in prior work to their capability to model supra-segmental temporal information (SST), i.e., learn rhythmic-prosodic characteristics of speech in addition to spectral features. In this paper, we (i) present and apply a novel test to quantify to what extent the performance of state-of-the-art neural networks for speaker recognition can be explained by modeling SST; and (ii) present several means to force respective nets to focus more on SST and evaluate their merits. We find that a variety of CNN- and RNN-based neural network architectures for speaker recognition do not model SST to any sufficient degree, even when forced. The results provide a highly relevant basis for impactful future research into better exploitation of the full speech signal and give insights into the inner workings of such networks, enhancing explainability of deep learning for speech technologies. • Literature explains speaker recognition in neural nets by modeling of voice dynmaics. • Diagnostic: We quantify how well deep learning models actually capture dynamics. • Observation: State-of-the-art deep nets do not model speaker prosody but ignore it. • Interpretation as "cheating": Achieving high without putting in due effort. • Outlook: Increasing task difficulty biases models towards prosody, but not enough. [ABSTRACT FROM AUTHOR]

Traditional sentiment analysis focuses on text-level sentiment mining, transforming sentiment mining into classification or regression problems, resulting in a sentiment analysis low accuracy rate. Sentiment analysis refers to the use of natural language processing, text analysis, and computational linguistics to systematically identify, extract, quantify, and study sentimental states. Therefore, more scholars have begun to focus on speech recognition and facial expression recognition research, and extracting and analysing people's sentiment tendencies can improve sentiment recognition accuracy. Traditional single-modal sentiment analysis can no longer meet people's needs. Therefore, this paper proposes a multimodal sentiment analysis method based on the multimodal sentiment analysis method that can obtain more sentimental information sources and help people make better decisions. The experimental results in this paper show that the highest recognition rates of CNN-SVM, RNN-SVM, and CRNN-SVM were 76.8%, 71.2%, and 93.5%, respectively. It can be seen that CRNN-SVM has the highest sentiment tendency recognition rate in deep learning, so it is suitable to apply CRNN-SVM to sentiment tendency analysis system design in this paper. The average accuracy rate of the system designed in this paper was 91%, and the stability was also very strong, which shows that the system designed in this paper is meaningful. The main contribution of this paper is based on the limitations of single-mode emotion analysis. It proposes a multimode emotion analysis method and introduces a convolutional neural network to help people obtain more emotional information sources to meet their needs. [ABSTRACT FROM AUTHOR]

This paper presents a new case study examining the use of police transcripts to assist the court in understanding poor-quality forensic audio admitted as evidence in criminal trials. The 1995 trial it studies was the first major Australian case to include extensive expert opinions about police transcripts provided by the prosecution. Despite the fact that experts on both sides noted serious problems with the police transcripts, the judge allowed them to assist the jury, with the expert opinions offered as (ineffectual) subsequent commentary. The legal procedures the judge used in doing this were upheld on appeal, and have been followed ever since as a model for judges admitting opinions of both police and experts. The paper demonstrates how these procedures (unintentionally) privileged the opinions of police "ad hoc" experts over those of genuine experts, enabling the erroneous transcripts to influence not only the 1995 verdict, but a 1997 appeal and a 2014 inquiry. Analysis reveals the reason for these anomalies as the fact that the procedures incorporate misconceptions about spoken language and its representation in a transcript, which, though they have been thoroughly refuted by linguistic science over many decades, remain deeply embedded in the "common knowledge" accepted by wider society – including powerful institutions such as the law. The paper ends by calling on Australian linguists to find effective ways to address the misconceptions that affect the legal handling of forensic audio, by building further on the success of other branches of forensic linguistics in seeking direct engagement with the judiciary outside the trial process. The first step in achieving this is for linguists to gain a thorough understanding of how the legal procedures for handling poor-quality forensic audio operate, both in principle and in practice. The aim of the present paper is to contribute to that understanding. [ABSTRACT FROM AUTHOR]

Qualitative researchers can usually discern the difference between obedient speech and fearless, critical, or oppositional speech. Yet the context in which speech acts are performed is necessarily uneven, such that the same people who might speak freely in one place are often quick to engage in obedient speech in another. Speech acts also depend on the speaker's positionality, meaning that some speakers may have the privilege to act as 'truth‐tellers' and speak freely, whereas the positionality of others does not enable this. This paper considers how these contextual factors can be overlooked when liberal speech norms are taken for granted. Engaging with Michel Foucault's writing on parrhesia, I highlight the issues of positionality and context in defining how socio‐political borders are drawn around free ('fearless') speech as opposed to obedient ('performative') speech. I show how parrhesia opens up key questions for qualitative research about the politicisation of free versus obedient speech through space and time. This paper examines the problem of liberal speech norms and context in qualitative research. Engaging with Michel Foucault's writing on parrhesia, it calls for more attention to a speaker's positionality and the context of speech, which intersect with normative practices of drawing borders between 'free speech' and 'obedient speech'. [ABSTRACT FROM AUTHOR]

Background: Phonetic transcription of disordered speech samples is especially crucial for the assessment and treatment of functional or organic speech‐sound disorders. Previous studies show that students who struggle with the identification and segmentation of speech sounds are more likely to encounter difficulties with clinical phonetic transcription. Aims: To clarify the connection between phonemic awareness and phonetic transcription in Mandarin‐speaking undergraduate students, as well as the improvement of these skills after phonetic transcription training. Methods & Procedures: A pretest–post‐test design was adopted for this study. The experimental group (n = 38) consisted of students majoring in audiology and speech–language pathology who were given 12 weeks of transcription training. The control group (n = 33) consisted of audiology and speech–language pathology majors who completed the same module a year earlier without undergoing the transcription training. Two instruments were used to assess their skills: (1) the Phonemic Awareness Skill Test (PA test) and (2) the Phonetic Transcription Skill Test (PT test) designed for the purposes of the present study. Outcomes & Results: Performance on most subtests of the PT and PA tests improved significantly at post‐test for the experiment group, although only PT training was provided. After phonetic transcription training, the relationship between phonemic addition abilities and phonetic transcription strengthened. Phonemic identification was found to be a predictor of undergraduate students' PT skills of disordered speech samples. Conclusions & Implications: Results from this study provide insights into the role of phonetic transcription training in improving phonemic awareness and phonetic transcription skills. What This Paper Adds: What is already known on the subject: Phonetic transcription of disordered speech samples is especially crucial for the assessment and treatment of functional or organic speech‐sound disorders. Phonemic awareness is not only an essential ability for reading alphabetic language but also associated with acquiring phonetic transcription skills. What this paper adds to existing knowledge: A dynamically changing association between phonemic transcription and phonetic transcription during transcription training was reported. Phonetic transcription training of disordered speech samples acted as a driving force for deeper relationships between the two skills. What are the potential or actual clinical implications of this work?: Training phonetic recognition and transcription equips future and practising speech–language pathologists with better skills in managing speech‐sound disorders. Implications for phonemic awareness training also have the potential to extend to the literacy skills of alphabetic language speakers. [ABSTRACT FROM AUTHOR]

Background: Patients with anomic aphasia experience difficulties in narrative processing. General discourse measures are time consuming and require necessary skills. Core lexicon analysis has been proposed as an effort‐saving approach but has not been developed in Mandarin discourse. Aims: This exploratory study was aimed (1) to apply core lexicon analysis in Mandarin patients with anomic aphasia at the discourse level and (2) to verify the problems with core words among people with anomic aphasia. Methods & Procedure: The core nouns and verbs were extracted from narrative language samples from 88 healthy participants. The production of core words for 12 anomic aphasia and 12 age‐ and education‐matched controls were then calculated and compared. The correlation between the percentages and the Aphasia Quotients of the revised Western Aphasia Battery was analyzed as well. Outcomes & Results: The core nouns and verbs were successfully extracted. Patients with anomic aphasia produced fewer core words than healthy people, and the percentages differed significantly in different tasks as well as word classes. There was no correlation between the core lexicon use and the severity of aphasia in patients with anomic aphasia. Conclusions & Implications: Core lexicon analysis may potentially serve as a clinician‐friendly manner of quantifying core words produced at the discourse level in Mandarin patients with anomic aphasia. WHAT THIS PAPER ADDS: What is already known on the subject: Discourse analyses in aphasia assessment and treatment have increasingly garnered attention. Core lexicon analysis based on English AphasiaBank has been reported in recent years. It is correlated with microlinguistic and macrolinguistic measures in aphasia narratives. Nevertheless, the application based on Mandarin AphasiaBank is still under development in healthy individuals and patients with anomic aphasia. What this paper adds to existing knowledge: A Mandarin core lexicon set was developed for different tasks. The feasibility of core lexicon analysis to evaluate the corpus of patients with anomic aphasia was preliminarily discussed and the speech performance of patients and healthy people was then compared to provide a reference for the evaluation and treatment of clinical aphasia corpus. What are the potential or actual clinical implications of this work?: The purpose of this exploratory study was to consider the potential use of core lexicon analysis to evaluate core word production in narrative discourse. Moreover, normative and aphasia data were provided for comparison to develop clinical use for Mandarin patients with anomic aphasia. [ABSTRACT FROM AUTHOR]

In less than three decades, the concept 'cerebellar neurocognition' has evolved from a mere afterthought to an entirely new and multifaceted area of neuroscientific research. A close interplay between three main strands of contemporary neuroscience induced a substantial modification of the traditional view of the cerebellum as a mere coordinator of autonomic and somatic motor functions. Indeed, the wealth of current evidence derived from detailed neuroanatomical investigations, functional neuroimaging studies with healthy subjects and patients and in-depth neuropsychological assessment of patients with cerebellar disorders shows that the cerebellum has a cardinal role to play in affective regulation, cognitive processing, and linguistic function. Although considerable progress has been made in models of cerebellar function, controversy remains regarding the exact role of the 'linguistic cerebellum' in a broad variety of nonmotor language processes. This consensus paper brings together a range of different viewpoints and opinions regarding the contribution of the cerebellum to language function. Recent developments and insights in the nonmotor modulatory role of the cerebellum in language and some related disorders will be discussed. The role of the cerebellum in speech and language perception, in motor speech planning including apraxia of speech, in verbal working memory, in phonological and semantic verbal fluency, in syntax processing, in the dynamics of language production, in reading and in writing will be addressed. In addition, the functional topography of the linguistic cerebellum and the contribution of the deep nuclei to linguistic function will be briefly discussed. As such, a framework for debate and discussion will be offered in this consensus paper. [ABSTRACT FROM AUTHOR]

Automatic speech recognition systems based on end-to-end models (E2E-ASRs) can achieve comparable performance to conventional ASR systems while reproducing all their essential parts automatically, from speech units to the language model. However, they hide the underlying perceptual processes modelled, if any, and they have lower adaptability to multiple application contexts, and, furthermore, they require powerful hardware and an extensive amount of training data. Model-explainability techniques can explore the internal dynamics of these ASR systems and possibly understand and explain the processes conducting to their decisions and outputs. Understanding these processes can help enhance ASR performance and reduce the required training data and hardware significantly. In this paper, we probe the internal dynamics of three E2E-ASRs pre-trained for English by building an acoustic-syllable boundary detector for Italian and Spanish based on the E2E-ASRs' internal encoding layer outputs. We demonstrate that the shallower E2E-ASR layers spontaneously form a rhythmic component correlated with prominent syllables, central in human speech processing. This finding highlights a parallel between the analysed E2E-ASRs and human speech recognition. Our results contribute to the body of knowledge by providing a human-explainable insight into behaviours encoded in popular E2E-ASR systems. [ABSTRACT FROM AUTHOR]

Jamming recognition is a significant prior step to achieving effective jamming suppression, and the precise results of the jamming recognition will be beneficial to anti-jamming decisions. However, as the electromagnetic environment becomes more complex, the received signals may contain both suppression jamming and deception jamming, which is more challenging for existing methods focused on a single kind of jamming. In this paper, a recognition method for compound jamming based on a dual-channel neural network and feature fusion is proposed. First, feature images of compound jamming are extracted by the short-time Fourier transform and the wavelet transform. Feature images are then employed as inputs for the proposed network. During parallel processing in dual-channel, the proposed network can adaptively extract and learn task-relevant features via the attention modules. Finally, the output features in dual-channel are fused in the fusion subnetwork. Compared with existing methods, the proposed method can yield better recognition performance with less inference time. Additionally, compared with existing fusion strategies, the fusion subnetwork can further improve the recognition performance under low jamming-to-noise ratio conditions. Results with the semi-measured datasets also verify the feasibility and generalization performance of the proposed method. [ABSTRACT FROM AUTHOR]

This paper addresses a joint training approach applied to a pipeline comprising speech enhancement (SE) and automatic speech recognition (ASR) models, where an acoustic tokenizer is included in the pipeline to leverage the linguistic information from the ASR model to the SE model. The acoustic tokenizer takes the outputs of the ASR encoder and provides a pseudo-label through K-means clustering. To transfer the linguistic information, represented by pseudo-labels, from the acoustic tokenizer to the SE model, a cluster-based pairwise contrastive (CBPC) loss function is proposed, which is a self-supervised contrastive loss function, and combined with an information noise contrastive estimation (infoNCE) loss function. This combined loss function prevents the SE model from overfitting to outlier samples and represents the pronunciation variability in samples with the same pseudo-label. The effectiveness of the proposed CBPC loss function is evaluated on a noisy LibriSpeech dataset by measuring both the speech quality scores and the word error rate (WER). The experimental results reveal that the proposed joint training approach using the described CBPC loss function achieves a lower WER than the conventional joint training approaches. In addition, it is demonstrated that the speech quality scores of the SE model trained using the proposed training approach are higher than those of the standalone-SE model and SE models trained using conventional joint training approaches. An ablation study is also conducted to investigate the effects of different combinations of loss functions on the speech quality scores and WER. Here, it is revealed that the proposed CBPC loss function combined with infoNCE contributes to a reduced WER and an increase in most of the speech quality scores. [ABSTRACT FROM AUTHOR]

This article describes experiments on speech segmentation using long short‐term memory recurrent neural networks. The main part of the paper deals with multi‐lingual and cross‐lingual segmentation, that is, it is performed on a language different from the one on which the model was trained. The experimental data involves large Czech, English, German, and Russian speech corpora designated for speech synthesis. For optimal multi‐lingual modeling, a compact phonetic alphabet was proposed by sharing and clustering phones of particular languages. Many experiments were performed exploring various experimental conditions and data combinations. We proposed a simple procedure that iteratively adapts the inaccurate default model to the new voice/language. The segmentation accuracy was evaluated by comparison with reference segmentation created by a well‐tuned hidden Markov model‐based framework with additional manual corrections. The resulting segmentation was also employed in a unit selection text‐to‐speech system. The generated speech quality was compared with the reference segmentation by a preference listening test. [ABSTRACT FROM AUTHOR]

Acoustic echo cancellation (AEC) methods aim to suppress the acoustic coupling for hands‐free speech communication. Traditional AEC works by identifying the acoustic impulse response using adaptive algorithms. With recent research advances, deep learning has become an attractive choice for AEC. This paper introduces a two‐stage bidirectional long short term memory (TS‐BLSTM) framework, incorporating multi‐head self‐attention mechanisms after each BLSTM block. This is aimed at better capturing contextual information and further enhancing ability of the model to handle complex acoustic scenarios. The BLSTM blocks are utilized to aggregate magnitude spectrum information, modelling both time and frequency dependencies. Additionally, dilation convolution is introduced to broaden the range of information in each convolution output. The magnitude decoder estimates a mask for the input, resulting in the generation of an estimated magnitude spectrum for near‐end speech. Experimental results indicate that the proposed method achieves promising outcomes. [ABSTRACT FROM AUTHOR]

People are curious about voice commands for the next generation of interaction. It will play a dominant role in communicating with smart devices in the future. However, language remains a significant barrier to the widespread use of these devices. Even the existing models for the traditional languages need to compute extensive parameters, resulting in higher computational costs. The most inconvenient in the latest advanced models is that they are unable to function on devices with constrained resources. This paper proposes a novel end-to-end speech recognition based on a low-cost Bidirectional Long Short Term Memory (BiLSTM) attention model. The mel-spectrogram of the speech signals has been generated to feed into the proposed neural attention model to classify isolated words. It consists of three convolution layers followed by two layers of BiLSTM that encode a vector of length 64 to get attention against the input sequence. The convolution layers characterize the relationship among the energy bins in the spectrogram. The BiLSTM network removes the prolonged reliance on the input sequence, and the attention block finds the most significant region in the input sequence, reducing the computational cost in the classification process. The encoded vector by the attention head is fed to three-layered fully connected networks for recognition. The model takes only 133K parameters, less than several current state-of-the-art models for isolated word recognition. Two datasets, the Speech Command Dataset (SCD), and a self-made dataset we developed for fifteen spoken colors in the Bengali dialect, are utilized in this study. Applying the proposed technique, the performance evaluation with validation and test accuracy in the Bengali color dataset reaches 98.82% and 98.95%, respectively, which outperforms the current state-of-the-art models regarding accuracy and model size. When the SCD has been trained using the same network model, the average test accuracy obtained is 96.95%. To underpin the proposed model, the outcome is compared with the recent state-of-the-art models, and the result shows the superiority of the proposed model. [ABSTRACT FROM AUTHOR]

Speech perception is influenced by language-specific phonological knowledge. While phonotactics has long been established to play a role, the study of how phonological alternations influence perception is still in its infancy. In this paper, we make a case for the latter by investigating the role of regressive voicing assimilation (RVA) in the perception of obstruent clusters in Emilian dialects of Italian. We provide empirical evidence from a phoneme-detection task, in which Emilian listeners reported to have heard [b] significantly more often in stimuli with a /p/ before a voiced obstruent (RVA context) than before a vowel (non-RVA context). Our experimental findings add to recent work on the influence of phonology on speech perception. In addition, we provide an explicit formalisation, which bolsters the need for a rigid distinction between phonetic, surface and underlying representation, and an explicit mapping between all three, both in the process of speech production and comprehension. [ABSTRACT FROM AUTHOR]

The purpose of the present study was to (1) translate and validate the PEACH + Rating Scale in Malay and (2) establish normative curves as a function of age and examine test-retest reliability for the Malay and English versions of PEACH+. This is a cross-sectional study that used a convenient sampling technique. One hundred and fifty-seven parents of typically developing children aged between 4 months and 7 years participated in the study. Forty-nine completed the Malay PEACH + in a pen-to-paper format (Aim 1). One hundred and eight parents completed PEACH + online (69 completed the Malay version and 39 the English version), and 20 of them completed the questionnaire twice (Aim 2). The PEACH + in Malay showed high internal consistency and item-total correlation. The normative data revealed that scores for frequency of auditory behaviour increased rapidly with age until about 20 months and reached an asymptote of around 90% by about 40 months of age. A similar trend was observed for ease of listening scores, which asymptoted around 85%. The validated Malay PEACH + Rating Scale can be used as a guide to monitor auditory functional performance and listening efforts of Malaysian children in real-world environments. [ABSTRACT FROM AUTHOR]

This paper presents an operational annotation system for (dis)fluencies in typical and atypical speech, based on existing standard annotation schemes previously established in the literature. Grounded in a functional approach to (dis)fluency, we address some of the conceptual and technical limitations found in previous annotation models, and offer an integrated and inclusive system which is compatible with different multi-layered annotation software such as Praat or ELAN. Our aim is twofold: to create comparable annotated corpora both in typical and atypical speech, and to provide natural language processing and the health sector with applications for diagnostic and therapy in speech disorders. [ABSTRACT FROM AUTHOR]

Background: The Chear open-set performance test (COPT), which uses a carrier phrase followed by a monosyllabic test word, is intended for clinical assessment of speech recognition, evaluation of hearing-device performance, and the fine-tuning of hearing devices for speakers of British English. This paper assesses practice effects, test–retest reliability, and the variability across lists of the COPT. Method: In experiment 1, 16 normal-hearing participants were tested using an initial version of the COPT, at three speech-to-noise ratios (SNRs). Experiment 2 used revised COPT lists, with items swapped between lists to reduce differences in difficulty across lists. In experiment 3, test–retest repeatability was assessed for stimuli presented in quiet, using 15 participants with sensorineural hearing loss. Results: After administration of a single practice list, no practice effects were evident. The critical difference between scores for two lists was about 2 words (out of 15) or 5 phonemes (out of 50). The mean estimated SNR required for 74% words correct was −0.56 dB, with a standard deviation across lists of 0.16 dB. For the participants with hearing loss tested in quiet, the critical difference between scores for two lists was about 3 words (out of 15) or 6 phonemes (out of 50). [ABSTRACT FROM AUTHOR]

In this paper, word length in the texts of public speeches by USA and UK politicians is analyzed. More than 300 speeches delivered over the past two hundred years were studied. It is found that the lognormal distribution better describes the distribution of word length than do the Weibull and Poisson distributions, for example. It is shown that the length of words does not change significantly over time (the average value either does not change or slightly decreases, and the mode slightly increases). These results are fundamentally different from those obtained previously for sentence lengths and indicate that, in terms of quantitative linguistic analysis, the word length in politicians' speech has not evolved over the last 200 years and does not obey the principle of least effort proposed by G. Zipf. [ABSTRACT FROM AUTHOR]

Background: Diadochokinetic (DDK) rate tasks are extensively used in the evaluation of speech disorders; however, it is unclear how the different types of speech stimuli affect DDK rate performance. Aims: To investigate the effect of age, gender and type of stimuli (non‐words versus real words) on the DDK rates in individuals across the lifespan and to provide normative data for Greek. Also to examine the discrimination ability of the speech DDK stimuli administered (non‐words and real words) based on a dual DDK assessment protocol using a polytomous item response theory (IRT) model. Methods & Procedure: The participants were 1747 monolingual Greek speakers (376 children, aged 4–17 years; and 1371 adults, aged 18–90+ years). All participants had normal hearing acuity which allowed them to understand and follow instructions. Participants with a medical condition or a language disorder which would affect DDK rate performance were excluded from the study. The time‐by‐count method was used, and all participants had to repeat as accurately and fast as possible: (1) four disyllabic non‐words (/'gaba/, /'taka/, /'kata/, /'baga/), (2) four disyllabic real words (/'kapa/, /'tapa/, /ka'la/, /'paka/) and (3) two trisyllabic non‐words (/'pataka/, /'badaga/). All responses were recorded and the speech samples that did not include at least 5 s of correct repetitions were excluded from the analysis. Outcomes & Results: Age affected significantly DDK rates with performance increasing gradually until approximately the age of 40 and then gradually decreasing. Gender had no effect. Overall, there was a significant advantage of disyllabic real word stimuli over disyllabic non‐word stimuli and of trisyllabic non‐word stimuli over disyllabic non‐word stimuli on DDK rates performance. IRT analysis suggested that the data fit the polytomous model reasonably well and all DDK stimuli (real words and non‐words) showed a strong relationship (loadings > 0.50) with the latent trait. Conclusions & Implications: The current study complements prior research which supports that age and type of stimuli significantly affect DDK rates performance. It is the first study, that testifies to the benefit of real‐word stimuli over non‐word stimuli on DDK rates across the lifespan in a large representative sample. The implementation of IRT analysis provides empirical evidence about the discrimination ability of the DDK stimuli administered and confirms the reliability of this dual DDK assessment protocol. These findings are valuable for clinicians who work with motor speech disorders. WHAT THIS PAPER ADDS: What is already known on this subject: Age, type of stimuli (real words versus non‐words) and language significantly affect DDK rates performance. Current research strongly suggests the administration of language‐specific norms since language‐dependent features seem to have a noteworthy effect on the DDK rates, but scarce evidence exists about the discriminatory ability of the DDK speech stimuli commonly administered. What this study adds to the existing knowledge: Conflicting findings have been reported about the effect of different types of DDK speech stimuli (real words and non‐words) but no study to date has evaluated their discriminatory abilities. The current study is the first to implement a polytomous IRT model to examine this issue. This is also the first study to attempt an investigation of the effect of types of stimuli (real words versus non‐words) on a large representative sample across the lifespan (4–90+ years) and to provide normative data for Greek. What are the practical and clinical implications of this work?: The present study offers concrete evidence about the advantage of real‐word stimuli over non‐word stimuli in Greek, as well as normative data for the Greek‐speaking populations. Moreover, the IRT analysis testifies to the discriminatory ability of real‐word and non‐word stimuli affirming the reliability of the present dual DDK assessment protocol as a psychometrically sound measure of DDK ability. The above has significant value for clinicians who work with individuals with motor speech disorders as the protocol can help them with the diagnosis and differential diagnosis of motor speech disorders. [ABSTRACT FROM AUTHOR]

In this study, novel Spectro-Temporal Energy Ratio features based on the formants of vowels, linearly spaced low-frequency, and logarithmically spaced high-frequency parts of the human auditory system are introduced to implement single- and cross-corpus speech emotion recognition experiments. Since the underlying dynamics and characteristics of speech recognition and speech emotion recognition differ too much, designing an emotion-recognition-specific filter bank is mandatory. The proposed features will formulate a novel filter bank strategy to construct 7 trapezoidal filter banks. These novel filter banks differ from Mel and Bark scales in shape and frequency regions and are targeted to generalize the feature space. Cross-corpus experimentation is a step forward in speech emotion recognition, but the researchers are usually chagrined at its results. Our goal is to create a feature set that is robust and resistant to cross-corporal variations using various feature selection algorithms. We will prove this by shrinking the dimension of the feature space from 6984 down to 128 while boosting the accuracy using SVM, RBM, and sVGG (small-VGG) classifiers. Although RBMs are considered no longer fashionable, we will show that they can achieve outstanding jobs when tuned properly. This paper discloses a striking 90.65% accuracy rate harnessing STER features on EmoDB. [ABSTRACT FROM AUTHOR]

This paper examines the perceptual threshold in patterns of tonal timing (alignment) of Falling versus Low tones. The results indicate a remarkable sensitivity among the listeners. In a perception experiment with 30 participants, we tested how native speakers of the West Nilotic language Nuer responded to stimuli in which the timing of the F0 fall that distinguishes Low versus Fall following a High target is manipulated. We measured the threshold for the responses to shift tone perception from 25% to 75%. The results show that listeners only needed an average of 19 ms to differentiate between the melodic shapes and as little as 13 ms for one item. Perceptual sensitivity this fine-grained is not expected based on what is known about the Just Noticeable Difference (JND) from previous studies. Results from non-tonal languages report a sensitivity threshold for tonal timing of at least 50 ms at category boundaries. This difference suggests that whether or not subjects speak a tone language may be a determining factor in their JND. [ABSTRACT FROM AUTHOR]

Background Music is an integral part of our lives and is often played in public places like restaurants. People exposed to music that contained alcohol-related lyrics in a bar scenario consumed significantly more alcohol than those exposed to music with less alcohol-related lyrics. Existing methods to quantify alcohol exposure in song lyrics have used manual annotation that is burdensome and time intensive. In this paper, we aim to build a deep learning algorithm (LYDIA) that can automatically detect and identify alcohol exposure and its context in song lyrics. Methods We identified 673 potentially alcohol-related words including brand names, urban slang, and beverage names. We collected all the lyrics from the Billboard's top-100 songs from 1959 to 2020 (N = 6110). We developed an annotation tool to annotate both the alcohol-relation of the word (alcohol, non-alcohol, or unsure) and the context (positive, negative, or neutral) of the word in the song lyrics. Results LYDIA achieved an accuracy of 86.6% in identifying the alcohol-relation of the word, and 72.9% in identifying its context. LYDIA can distinguish with an accuracy of 97.24% between the words that have positive and negative relation to alcohol; and with an accuracy of 98.37% between the positive and negative context. Conclusion LYDIA can automatically identify alcohol exposure and its context in song lyrics, which will allow for the swift analysis of future lyrics and can be used to help raise awareness about the amount of alcohol in music. Highlights Developed a deep learning algorithm (LYDIA) to identify alcohol words in songs. LYDIA achieved an accuracy of 86.6% in identifying alcohol-relation of the words. LYDIA's accuracy in identifying positive, negative, or neutral context was 72.9%. LYDIA can automatically provide evidence of alcohol in millions of songs. This can raise awareness of harms of listening to songs with alcohol words. [ABSTRACT FROM AUTHOR]

Internet, also known as the Internet, refers to the huge Internet connected between the LAN and the LAN, which connects a huge Internet from a set of common protocols. Today, accurate real-time multimedia information publishing technology has been widely used in many fields. In order to achieve the purpose of information disclosure, the multimedia information disclosure system uses multimedia resources as an object, and the resources are displayed by the user's editing of the management party multimedia resource. With the increased workload of management and multimedia's maintenance information, the current multimedia information management and control software cannot meet the needs of the market. In order to ensure the real-time, accuracy of the multimedia information disclosure, it is necessary to check if the speech recognition technology can accurately collect voice information. Techniques are particularly important for speech recognition, so it is also called automatic voice recognition; its purpose is to convert the words content contained in human speech to computer recognition content. It is different from the speaker's identification, the speaker is different, the latter tries to identify or confirm that the person who makes a voice, rather than identifying or confirming the vocabulary included. However, voice recognition technology requires a very perfect system design, and the system design can use system science ideas and methods, according to systematic analysis results, it is also possible to design a new system. Therefore, this paper uses the voice system on the Internet to provide a solution to the inaccuracicity in speech recognition technology, and use speech recognition techniques to disclose multimedia information. [ABSTRACT FROM AUTHOR]

Drug addiction can cause severe damage to the human brain, leading to significant problems in cognitive processing, such as irritability, speech distortions, and exaggeration of negative stimuli. Speech plays a fundamental role in social interaction, including both the production and perception. The ability to perceive communicative functions conveyed through speech is crucial for successful interpersonal communication and the maintaining good social relationships. However, due to the limited number of previous studies, it remains unclear whether the cognitive disorder caused by drug addiction affects the perception of communicative function conveyed in Mandarin speech. To address this question, we conducted a perception experiment involving sixty male participants, including 25 heroin addicts and 35 healthy controls. The experiment aimed to examine the perception of three communicative functions (i.e., statement, interrogative, and imperative) under three background noise conditions (i.e., no noise, SNR [Signal to Noise Ratio] = 10, and SNR = 0). Eight target sentences were first recorded by two native Mandarin speakers for each of the three communicative functions. Each half was then combined with Gaussian White Noise under two background noise conditions (i.e., SNR = 10 and SNR = 0). Finally, 48 speech stimuli were included in the experiment with four options provided for perceptual judgment. The results showed that, under the three noise conditions, the average perceptual accuracies of the three communicative functions were 80.66% and 38% for the control group and the heroin addicts, respectively. Significant differences were found in the perception of the three communicative functions between the control group and the heroin addicts under the three noise conditions, except for the recognition of imperative under strong noise condition (i.e., SNR = 0). Moreover, heroin addicts showed good accuracy (around 50%) in recognizing imperative and poor accuracy (i.e., lower than the chance level) in recognizing interrogative. This paper not only fills the research gap in the perception of communicative functions in Mandarin speech among drug addicts but also enhances the understanding of the effects of drugs on speech perception and provides a foundation for the speech rehabilitation of drug addicts. [ABSTRACT FROM AUTHOR]

Hearing speech in a particular voice is a common experience in spoken linguistic communication. Nevertheless, the nature and role(s) of that experience are relatively under-theorized in philosophy. What can we learn about auditory experiences of listening to speech in a voice from illusions? In this paper I discuss two illusions and two special effects in speech sound and voice perception: (1) the temporal induction illusion in speech, (2) the phantom words illusion, (3) the McGurk effect, and (4) the audiovisual experience of the voice-over translation. The first two phenomena support widely accepted claims of top-down influences in speech perception. The latter two illustrate the nature and scope of multimodal integration in speech and voice perception. My discussion of the voice-over translation technique and the experiences it may give rise to will expand the scope of the phenomena that are typically discussed in this area. [ABSTRACT FROM AUTHOR]

This paper presents the results of a listening experiment designed to assess annoyance and perceived loudness (PL) for several unmanned aircraft system (UAS) operations, with the listener simulated in indoor and outdoor positions. This research investigated (i) how participant responses change depending on UAS operation, (ii) which broadband metrics are most suitable for representing annoyance and PL, (iii) differences in noise level required to result in equal participant responses to different operations, and (iv) which sound quality metrics (SQMs) are significant for UAS noise perception. Results indicate annoyance and PL responses were greatest for landing operations with flyovers being the least annoying or loud. LAeq, LASmax, and loudness (N5) were the strongest predictors in representing annoyance. Offset analysis predicted small differences in annoyance responses between flyovers and other operations, but also indicated that flyovers would require an increase to LASmax of 3.3 to 6.3 dB compared to other operations to achieve equal PL. Loudness was the most significant SQM, with minor contributions from impulsivity for annoyance and PL when outside, and tonality for PL when indoors. These findings contribute to the understanding of UAS noise perception for the development of metrics and assessment methods accounting for the characteristics of UAS operations. [ABSTRACT FROM AUTHOR]

In this paper, we study the speech emotion feature optimization using stochastic optimization algorithms, and feature compensation using deep neural networks. We also proposed to use accentuation-based fusion for long-time speech emotion recognition. Firstly, the extraction method of emotional features is studied, and a series of speech features are constructed for the recognition of emotion. Secondly, we propose a method of sample adaptation through denoising autoencoder to enhance the versatility of features through the mapping of sample features to improve adaptive ability. Thirdly, GA and SFLA are used to optimize the combination of features to improve the emotion recognition results at the utterance level. Finally, we use transformer model to implement accentuation-based emotion fusion in long-time speech. The continuous long-time speech corpus, as well as the public available EMO-DB, are used for experiments. Results show that the proposed method can effectively improve the performance of long-time speech emotion recognition. [ABSTRACT FROM AUTHOR]

Automatic speech recognition (ASR) is a typical pattern recognition technology that converts human speeches into texts. With the aid of advanced deep learning models, the performance of speech recognition is significantly improved. Especially, the emerging Audio–Visual Speech Recognition (AVSR) methods achieve satisfactory performance by combining audio-modal and visual-modal information. However, various complex environments, especially noises, limit the effectiveness of existing methods. In response to the noisy problem, in this paper, we propose a novel cross-modal audio–visual speech recognition model, named CATNet. First, we devise a cross-modal bidirectional fusion model to analyze the close relationship between audio and visual modalities. Second, we propose an audio–visual dual-modal network to preprocess audio and visual information, extract significant features and filter redundant noises. The experimental results demonstrate the effectiveness of CATNet, which achieves excellent WER, CER and converges speeds, outperforms other benchmark models and overcomes the challenge posed by noisy environments. • Proposing a novel cross-modal audio–visual speech recognition network, named CATNet. • Devising a cross-modal bidirectional fusion model. • Devising an audio–visual dual-modal speech recognition network. • CATNet is robust against noises and outperforms other benchmarks. [ABSTRACT FROM AUTHOR]

Children with single-sided deafness (SSD) may experience delays in language and speech development. Reduced speech discrimination and poor sound localization abilities in young SSD patients may result in greater cognitive efforts required to focus and process auditory information, as well as increased listening-related fatigue. Consequently, these children can have a higher risk of academic failure and are often in need of extra help at school. Recently, cochlear implants (CIs) have been introduced as a rehabilitative option for these children, but their effectiveness is still a topic of debate. A literature review was performed according to PRISMA guidelines, searching the Medline database from inception to October 2023. The research identified nine papers that met the inclusion criteria. Data extracted from the selected studies included 311 children affected by SSD and cochlear implants. The reported audiological outcomes were further analyzed. Overall, a high level of satisfaction was described by parents of children with SSD and CI, and those who received a CI under the age of 3 presented better results. However, a proportion of patients did not use the device daily. Our review highlights the possible, and still controversial, role of CI for the hearing rehabilitation of children with unilateral deafness, underlining the need for further research in this field. To date, careful and comprehensive counseling with the child and the family is necessary before considering this option. [ABSTRACT FROM AUTHOR]