Search

IntroductionObjectively predicting speech intelligibility is important in both telecommunication and human-machine interaction systems. The classic method relies on signal-to-noise ratios (SNR) to successfully predict speech intelligibility. One exception is clear speech, in which a talker intentionally articulates as if speaking to someone who has hearing loss or is from a different language background. As a result, at the same SNR, clear speech produces higher intelligibility than conversational speech. Despite numerous efforts, no objective metric can successfully predict the clear speech benefit at the sentence level.MethodsWe proposed a Syllable-Rate-Adjusted-Modulation (SRAM) index to predict the intelligibility of clear and conversational speech. The SRAM used as short as 1 s speech and estimated its modulation power above the syllable rate. We compared SRAM with three reference metrics: envelope-regression-based speech transmission index (ER-STI), hearing-aid speech perception index version 2 (HASPI-v2) and short-time objective intelligibility (STOI), and five automatic speech recognition systems: Amazon Transcribe, Microsoft Azure Speech-To-Text, Google Speech-To-Text, wav2vec2 and Whisper.ResultsSRAM outperformed the three reference metrics (ER-STI, HASPI-v2 and STOI) and the five automatic speech recognition systems. Additionally, we demonstrated the important role of syllable rate in predicting speech intelligibility by comparing SRAM with the total modulation power (TMP) that was not adjusted by the syllable rate.DiscussionSRAM can potentially help understand the characteristics of clear speech, screen speech materials with high intelligibility, and convert conversational speech into clear speech.

End-to-end speech translation (ST) has attracted substantial attention due to its less error accumulation and lower latency. Based on triplet ST data ⟨ speech-transcription-translation ⟩ , multitask learning (MTL) that utilizes machine translation ⟨ transcription-translation ⟩ or automatic speech recognition ⟨ speech-transcription ⟩ task to assist in training ST model is widely employed. However, current MTL methods often suffer from subnet role mismatch, semantic inconsistency, or usually focus only on transferring knowledge from automatic speech recognition (ASR) or machine translation (MT) task, leading to insufficient transferring of cross-task knowledge. To solve these problems, we propose the multitask co-training network (MCTN) to jointly model ST, MT, and ASR tasks. Specifically, the ASR task enables the acoustic encoder to better capture local information of speech frames, and the MT task enhances the translation capability of the model. MCTN benefits from three key aspects: a well-designed multitask framework to fully exploit the association between tasks, a model decoupling and parameter sharing method to maintain consistency in subnet roles, and a co-training strategy to utilize task information in triplet ST data. Our experiments show that MCTN achieves state-of-the-art results, when using only MuST-C dataset, and significantly outperforms strong end-to-end ST baselines and cascaded systems when external data are available. [ABSTRACT FROM AUTHOR]

The classification of musical emotion is a challenging problem due to the subjectivity of human perception. In order to analyse the emotion contained in music from the perspective of product supply chain management, the psychological model and characteristics of music emotion are introduced. Additionally, the current product supply chain and the composition of the two contracts are analysed based on the characteristics of online music products. Finally, a speech emotion recognition model based on a Fuzzy Cognitive Map (FCM) is proposed. Two corpora containing five emotions (sadness, anger, happiness, surprise, and boredom) are selected to test the performance and characteristics of the network. In order to evaluate the system performance, different acoustic features are extracted, including prosodic features, formant features, marginal spectral features, and Mel‐Frequency Cepstral Coefficients (MFCC). The results show that the average recognition rate of MFCC features on the Berlin emotional speech database is 70.36%, which is 0.75% higher than the second‐highest edge spectrum features and achieves better results. Meanwhile, on this database, the Musical Emotion Recognition System based on FCM (MERS‐FCM) is improved by 20.36%, 7.34%, and 4.12% compared with Back Propagation, K‐nearest neighbour, and support vector machine, respectively, which shows that MERS‐FCM has better performance. The research of the music emotion recognition system based on FCM can provide more reference paths for the use and development of music emotion information. [ABSTRACT FROM AUTHOR]

Developing an automatic speaker verification (ASV) system for children's speech presents significant challenges. One major obstacle is the scarcity of domain-specific data. This issue is exacerbated when dealing with short speech utterances, a relatively unexplored area in children's ASV. Voice biometric systems struggle during enrollment and verification phase, when faced with inadequate speech data, both in volume as well as in duration. To address data scarcity, this paper explores various in-domain and out-of-domain data augmentation techniques. Out-of-domain data from adult speakers, which have distinct acoustic attributes from children, are modified using techniques like voice-conversion, prosody and formant modification to make them acoustically similar to children's speech. In-domain data augmentation involves perturbing the speed of children's speech. This combined data augmentation approach not only increases training data volume but also captures missing target attributes, resulting in a significant 43.91% reduction in equal error rate (EER) compared to the baseline system. Additionally, the paper addresses the challenge of preserving higher-frequency components in children's speech. It achieves this by concatenating conventional Mel-frequency cepstral coefficients (MFCC) with Inverse-Mel-frequency cepstral coefficient (IMFCC) features at the frame level. The low canonical correlation between MFCC and IMFCC feature vectors motivates this fusion. The feature concatenation approach, when combined with proposed data augmentation, results in an appreciable reduction of 48.51% in the overall EER, demonstrating its effectiveness in improving the performance of children's ASV system. [ABSTRACT FROM AUTHOR]

Speech presence probability (SPP) and gain functions such as Wiener filter or MMSE estimators require an estimate of the a-priori signal-to-noise ratio (SNR). However, the estimation of the a-priori SNR is computationally involved and sensitive to noise variations. This paper proposes to approximate the SPP and the overall gain function of a speech enhancement system by using sigmoid functions to reduce the need of estimating the a-prior SNR. By applying an approximation via the sigmoid functions it is shown that only the a-posteriori estimate of SNR is needed, resulting in a low complexity system. The sigmoid function is designed with an optimization algorithm to optimize its parameters with respect to speech quality measures. The optimization algorithm is based on the idea that the solution obtained for a given problem should move towards the best solution and avoid the worst solution. The proposed algorithm requires minimal control parameters and does not require any algorithm specific parameters. Simulation results show that the proposed sigmoid functions achieve good results in terms of speech quality measures when compared with existing methods while providing significantly lower complexity for implementation. [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]

Why do we prefer some singers to others? We investigated how much singing voice preferences can be traced back to objective features of the stimuli. To do so, we asked participants to rate short excerpts of singing performances in terms of how much they liked them as well as in terms of 10 perceptual attributes (e.g.: pitch accuracy, tempo, breathiness). We modeled liking ratings based on these perceptual ratings, as well as based on acoustic features and low-level features derived from Music Information Retrieval (MIR). Mean liking ratings for each stimulus were highly correlated between Experiments 1 (online, US-based participants) and 2 (in the lab, German participants), suggesting a role for attributes of the stimuli in grounding average preferences. We show that acoustic and MIR features barely explain any variance in liking ratings; in contrast, perceptual features of the voices achieved around 43% of prediction. Inter-rater agreement in liking and perceptual ratings was low, indicating substantial (and unsurprising) individual differences in participants' preferences and perception of the stimuli. Our results indicate that singing voice preferences are not grounded in acoustic attributes of the voices per se, but in how these features are perceptually interpreted by listeners. [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]

Neuromorphic speech recognition systems that use spiking neural networks (SNNs) and memristors are progressing in hardware development. The conventional manual preprocessing of audio signals is shifting toward event-based recognition with convolutional SNNs. Despite achieving high accuracy in classification, the efficient extraction of spatiotemporal features from audio events continues to be a substantial challenge. In this study, we introduce dynamic time-surface neurons (DTSNs) using volatile memristors featuring an adjustable temporal kernel decay, enabled by series-connected transistors with an Au/LiCoO2/Au configuration. DTSNs act as feature descriptors, enhancing the spatiotemporal feature extraction from event audio data. A two-layer SNN classifier, fully connected and incorporating a 1T1R nonvolatile memristor array, is trained to recognize the spatiotemporal features of the audio data. Our findings show classification accuracies of up to 95.91%, substantial improvements in computational efficiency, and increased noise resilience, confirming the promise of our memristor-based speech recognition system for practical applications. [ABSTRACT FROM AUTHOR]

According to the Universal Design for Learning (UDL) framework, as a textbased alternative to auditory information in videos or presentations, captions can make the content equally accessible, and multilingual subtitles can promote a cross-linguistic understanding of the content. We conducted a phenomenological study to understand the common meaning of the participants' experiences when using real-time automated captions/subtitles during live online class presentations. Twenty-four remote student participants were placed in three study groups. All participants were fluent in spoken and written English, eight could read in one or more additional languages, and none had a hearing disability. We used Microsoft PowerPoint Present Live via Zoom to deliver the online presentation to each group with real-time automated captions/subtitles, and then we conducted a focus group session with each group. Ten themes emerged and were clustered into three overarching themes: challenges, benefits, and interactions with subtitles. Overall, participants described a positive experience, perceiving the captions/subtitles as useful and accurate. Participants found the tool easy to use and highlighted the benefits of using captions/subtitles, such as providing access to live instruction for a wide audience and reinforcing learning for diverse student types. While they were able to troubleshoot connectivity and technological issues encountered, they experienced an apparent split-attention effect and noted limitations in the tool's inability to recognize different dialects. Findings contribute to educational research related to accessible live instruction in multilingual settings and could aid educators in selecting and integrating tools with real-time captioning/subtitling, in line with the UDL guidelines. [ABSTRACT FROM AUTHOR]

Emotion recognition systems from speech signals are realized with the help of acoustic or spectral features. Acoustic analysis is the extraction of digital features from speech files using digital signal processing methods. Another method is the analysis of time-frequency images of speech using image processing. The size of the features obtained by acoustic analysis is in the thousands. Therefore, classification complexity increases and causes variation in classification accuracy. In feature selection, features unrelated to emotions are extracted from the feature space and are expected to contribute to the classifier performance. Traditional feature selection methods are mostly based on statistical analysis. Another feature selection method is the use of metaheuristic algorithms to detect and remove irrelevant features from the feature set. In this study, we compare the performance of metaheuristic feature selection algorithms for speech emotion recognition. For this purpose, a comparative analysis was performed on four different datasets, eight metaheuristics and three different classifiers. The results of the analysis show that the classification accuracy increases when the feature size is reduced. For all datasets, the highest accuracy was achieved with the support vector machine. The highest accuracy for the EMO-DB, EMOVA, eNTERFACE'05 and SAVEE datasets is 88.1%, 73.8%, 73.3% and 75.7%, respectively. [ABSTRACT FROM AUTHOR]

Recent years have seen the advent of conversational humanoid robots. Implementing multiple robots is a promising approach for continuing human–robot conversations when a robot encounters speech-recognition errors. This strategy is effective for the user's verbal responses; however, people generally use nonverbal responses, such as nods or smiles, to respond to the interlocutor. In this study, we proposed a conversational strategy for twin robots in which the second robot recognized a human's nonverbal responses and interrupted the conversation by mentioning these responses. Moreover, we developed an interrogative dialogue system using simple nonverbal recognition modules. To verify the effectiveness of this conversational strategy, a subject-based experiment was conducted in which humans conversed with the two robots, and their impressions were evaluated through a questionnaire. We compared the outcomes under two conditions: whether the second robot's interruption was inserted or otherwise. Our results indicate that language-only ineffectiveness in responding to ambiguous responses regarding satisfaction and comprehension was mitigated. Thus, such a dialogue system is advantageous because it can facilitate robust conversation without relying solely on speech recognition. [ABSTRACT FROM AUTHOR]

To identify and describe the aims, methodological approaches, and major findings of studies on the use of STT among secondary pupils (age 12–18) with learning difficulties published from January 2000 to April 2022. This scoping review includes empirical studies published in peer-reviewed journals and grey literature between January 2000 and April 2022. Searches were conducted in April 2022 in three databases: ERIC, PsycINFO and Scopus. In addition, related reviews were manually screened for relevant papers. Eight peer-revied studies and five publications of grey literature were found to meet the inclusion criteria; two studies employed experimental designs, four employed quasi-experimental designs and seven employed explorative designs. Six studies described STT as an assistive technology (a compensatory aid for poor writing performance); two assessed STT as an instructional technology to determine whether it improves overall writing and related skills (e.g., reading). Results suggest that STT may increase pupils' abilities to produce texts with fewer errors, provide help with spelling and improve reading comprehension and word recognition. To date, there is a paucity of high-quality research on the use of STT among adolescents with LD. The scoping review shows that very little research has been conducted on the use of STT for adolescents with learning difficulties in secondary education. Findings from the studies identified five areas of interest: writing related skills, text assessment, writing processes, accuracy of the technology, and participants' experiences. Findings indicate that writing performance among students with learning difficulties improves when using STT. Parents, teachers, and pupils report positive experiences with the technology, particularly for students with severe reading and writing difficulties. There is a great need for more robust research on the use of speech-to-text technology (STT) in educational settings, especially on its effect on writing skills Studies describe STT as either an assistive (a compensatory aid for poor writing performance) or instructional technology (aiming to improve learning in general). It is important that practitioners are aware of the different aims and possible consequences of introducing STT to learners with writing difficulties. STT provides both opportunities and challenges for writers with learning difficulties in secondary education. Findings indicate that writing performance among students with learning difficulties improves when using STT, yet inaccuracy of the technology was presented as one of the main challenges. Parents, teachers, and pupils report positive experiences with the technology, particularly for students with severe reading and writing difficulties. [ABSTRACT FROM AUTHOR]

Spherical micro/nanorobots are reported as attractive candidates in nano‐surgery, contamination degradation, analytical chemistry, medical engineering, etc. Unlike macroscopic robots, the standard fabrication and inspection strategy to control the high quality of microrobots are rarely explored before. Overcoming the geometrical discrepancy is now essential to keep an identical locomotion behavior for spherical microrobot individuals for complex usages. Herein, acoustic‐based microbubble grippers are proposed controlled by a machine vision system to realize agile recognition and sorting of unqualified spherical microrobot models. Theoretical analysis and numerical simulation reveal the basic grabbing mechanisms via acoustic stimuli. The acoustic resonation of microbubbles can generate a locally enhanced acoustic field, attracting microrobots toward the bubble surface by the acoustic forces. The viscosity of microfluid makes it possible to instantaneously grab or release just by controlling the microbubble activation. Further, a machine vision‐based monitor and control system is proposed to enable automatic control for the entire process. Based on extracted characteristic parameters, target microrobot models are quickly selected and grabbed out. It is expected that such a combination of microgrippers with an advanced control system will pave the way for high‐qualified inspection for microrobots and improve their unique versatility in myriad micromanipulation scenarios. [ABSTRACT FROM AUTHOR]

Objectives: Detecting ablation site recurrence (ASR) after thermal ablation remains a challenge for radiologists due to the similarity between tumor recurrence and post-ablative changes. Radiomic analysis and machine learning methods may show additional value in addressing this challenge. The present study primarily sought to determine the efficacy of radiomic analysis in detecting ASR on follow-up computed tomography (CT) scans. The second aim was to develop a visualization tool capable of emphasizing regions of ASR between follow-up scans in individual patients. Materials and methods: Lasso regression and Extreme Gradient Boosting (XGBoost) classifiers were employed for modeling radiomic features extracted from regions of interest delineated by two radiologists. A leave-one-out test (LOOT) was utilized for performance evaluation. A visualization method, creating difference heatmaps (diff-maps) between two follow-up scans, was developed to emphasize regions of growth and thereby highlighting potential ASR. Results: A total of 55 patients, including 20 with and 35 without ASR, were included in the radiomic analysis. The best performing model was achieved by Lasso regression tested with the LOOT approach, reaching an area under the curve (AUC) of 0.97 and an accuracy of 92.73%. The XGBoost classifier demonstrated better performance when trained with all extracted radiomic features than without feature selection, achieving an AUC of 0.93 and an accuracy of 89.09%. The diff-maps correctly highlighted post-ablative liver tumor recurrence in all patients. Conclusions: Machine learning-based radiomic analysis and growth visualization proved effective in detecting ablation site recurrence on follow-up CT scans. [ABSTRACT FROM AUTHOR]

Dysarthria is a common manifestation across cerebellar ataxias leading to impairments in communication, reduced social connections, and decreased quality of life. While dysarthria symptoms may be present in other neurological conditions, ataxic dysarthria is a perceptually distinct motor speech disorder, with the most prominent characteristics being articulation and prosody abnormalities along with distorted vowels. We hypothesized that uncertainty of vowel predictions by an automatic speech recognition system can capture speech changes present in cerebellar ataxia. Speech of participants with ataxia (N=61) and healthy controls (N=25) was recorded during the "picture description" task. Additionally, participants' dysarthric speech and ataxia severity were assessed on a Brief Ataxia Rating Scale (BARS). Eight participants with ataxia had speech and BARS data at two timepoints. A neural network trained for phoneme prediction was applied to speech recordings. Average entropy of vowel tokens predictions (AVE) was computed for each participant's recording, together with mean pitch and intensity standard deviations (MPSD and MISD) in the vowel segments. AVE and MISD demonstrated associations with BARS speech score (Spearman's rho=0.45 and 0.51), and AVE demonstrated associations with BARS total (rho=0.39). In the longitudinal cohort, Wilcoxon pairwise signed rank test demonstrated an increase in BARS total and AVE, while BARS speech and acoustic measures did not significantly increase. Relationship of AVE to both BARS speech and BARS total, as well as the ability to capture disease progression even in absence of measured speech decline, indicates the potential of AVE as a digital biomarker for cerebellar ataxia. [ABSTRACT FROM AUTHOR]

Background: Tele‐collaborative projects serve as invaluable platforms for students from diverse countries to engage in cross‐cultural communication and exchange cultural knowledge. These projects offer immense benefits in terms of fostering intercultural competence among participants. However, one challenge arises when participants engage in spoken communication rather than written. Those who may not share a common first language often rely on a lingua franca such as English for intercultural communication. In such spoken exchanges, varying levels of language proficiency can lead to communication errors that hinder the intercultural learning process and impede the development of intercultural competence in their fellow participants. Objectives: In the present trial study, our primary objective was to tackle the language‐related challenge associated with spoken communication head‐on. Our focus centred on addressing the discrepancies arising in spoken communication, and we aimed to mitigate these discrepancies by utilizing speech‐enabled corrective feedback technology. Methods: In a tele‐collaborative project involving 28 graduate students from the People's Republic of China and Republic of Kazakhstan, the participants engaged in spoken communication and cultural exchange in English via the internet. Given the emphasis on spoken communication and the recognition that English was not their first language, we implemented a speech‐enabled corrective feedback (SECF) technology to ensure effective communication in terms of pronunciation, grammar, and word choice. The SECF technology detected and provided instant feedback on any speech mistakes, enabling participants to make corrections accordingly. The main aim of our study was to explore the potential benefits of integrating SECF technology into a tele‐collaborative project for the development of participants' intercultural competence. Additionally, we sought to gather insights into the students' perceptions of the SECF support. Results and Conclusions: The results of our study revealed significant improvements in the participants' intercultural competence across dimensions such as knowledge, skills, attitude, and awareness. Furthermore, the participants expressed positive perceptions of the SECF technology. Drawing from our findings, we offer valuable implications and recommendations for educators and researchers in the field, enhancing their understanding of the role of SECF technology in supporting intercultural competence development in tele‐collaborative settings. Lay Description: What is currently known about the subject matter: Tele‐collaborative projects are widely utilized to enhance intercultural competence.Low language proficiency of participants often poses a significant challenge in tele‐collaborative projects. What this paper adds: The present study addressed the language barrier by incorporating speech‐enabled corrective feedback (SECF) technology to support our tele‐collaborative project.The study explored how our tele‐collaborative project, supported by SECF, facilitated the development of intercultural competence. The implications of study findings for practitioners: Scholars and practitioners should consider implementing SECF technology to address language‐related barriers in tele‐collaborative projects.Guidelines based on our study can be utilized to design tele‐collaborative projects with SECF support.Training on technology and constant support needs to be provided to participants. [ABSTRACT FROM AUTHOR]

Machine learning techniques have proven to be a useful tool in cognitive neuroscience. However, their implementation in scalp‐recorded electroencephalography (EEG) is relatively limited. To address this, we present three analyses using data from a previous study that examined event‐related potential (ERP) responses to a wide range of naturally‐produced speech sounds. First, we explore which features of the EEG signal best maximize machine learning accuracy for a voicing distinction, using a support vector machine (SVM). We manipulate three dimensions of the EEG signal as input to the SVM: number of trials averaged, number of time points averaged, and polynomial fit. We discuss the trade‐offs in using different feature sets and offer some recommendations for researchers using machine learning. Next, we use SVMs to classify specific pairs of phonemes, finding that we can detect differences in the EEG signal that are not otherwise detectable using conventional ERP analyses. Finally, we characterize the timecourse of phonetic feature decoding across three phonological dimensions (voicing, manner of articulation, and place of articulation), and find that voicing and manner are decodable from neural activity, whereas place of articulation is not. This set of analyses addresses both practical considerations in the application of machine learning to EEG, particularly for speech studies, and also sheds light on current issues regarding the nature of perceptual representations of speech. This work is a methodological contribution to the growing field of applying machine learning techniques to EEG data. We compare different approaches to EEG decoding and offer recommendations for best practices to researchers who are interested in using classification analyses. Then, we demonstrate how these techniques can be applied to questions in spoken language processing, providing theoretical insights into the nature of perceptual representations of speech that conventional analysis methods have been unable to address. [ABSTRACT FROM AUTHOR]

At present, there are several communication tools employed to express human emotions. Among the numerous modes of communication, speech is the most predominant one for communicating with people effectively and efficiently. Speech emotion recognition (SER) plays a significant role in several signal processing applications. However, in both feature selection (FS) methods and reliable classifiers, determining their appropriate features has emerged as challenges in identifying the emotions expressed in Indian regional languages. In this work, a novel SER framework has been proposed to classify different speech emotions. Primarily, the proposed framework utilizes a preprocessing phase so as to alleviate the background noise and the artifacts present in input speech signal. Later on, the two new speech attributes related to energy and phase have been integrated with state-of-the-art attributes for examining speech emotion characteristics. The threshold-based feature selection (TFS) algorithm has been introduced to determine the optimal features by applying a statistical approach. An Indian regional language called Tamil Emotional dataset has been created for examining the proposed framework with the aid of standard machine learning and deep learning classifiers. The proposed TFS technique has been more suitable for Indian regional languages since it exhibits a superior performance with 97.96% accuracy compared to Indian English and Malayalam datasets. [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]

This paper proposes a dual-channel speech separation method based on spatial estimation via sparse Bayesian inference (SBI) and nonnegative matrix factorization (NMF). The spatial information estimated by traditional compressed sensing (CS) models is insufficient when two microphones receive limited columns of mixed signals. Considering the sparsity of peak values in the cross-correlation spectrum between two received signals, the proposed method builds a new CS model based on cross-correlation spectrum and applies SBI algorithm to solve this model to improve the estimation accuracy of spatial information. Combined the spatial information with the spectral features decomposed by NMF, NMF coefficient matrix masks belonging to individual source are generated for pre-separation. To mitigate retained potential interference components, a post-separation processing stage is designed using an expectation maximization (EM) algorithm based on a Gaussian mixture model (GMM). The estimated spatial information and binary time–frequency masks are used for parameter initialization of the EM algorithm. The experimental results using real-world speech data show that the proposed method can achieve better separation performance compared to various existing methods. [ABSTRACT FROM AUTHOR]

This work investigates the significance of Hilbert domain characterization of wavelet packets in classifying different emotion of speech signal. The goal of this paper is to create a new emotional speech database and introduce a new feature extraction approach that can recognize various emotions. The proposed feature, wavelet cepstral coefficients (WCC) are based on Hilbert spectrum analysis of the wavelet packet of the speech signal. The speaker-independent machine learning models are developed using multiclass support vector machine (SVM) and k-nearest neighbourhood (KNN) classifier. The approach is tested with newly developed Telugu Indian database and the EMOVO (Italian emotional speech) database. Our proposed wavelet features achieve a peak accuracy of 73.5%, further boosted by NCA feature selection by 3–5%, resulting in an improved unweighted average recall (UAR) of 78% for database 1 and 87.50% for database 2, employing optimal wavelet features in conjunction with SVM classification. The proposed features outperformed the baseline Mel-frequency cepstral coefficients (MFCC) feature. The performance of newly formulated features is better than other existing methodologies tested with different language databases. [ABSTRACT FROM AUTHOR]

Ultrasensitive transcriptional switches enable sharp transitions between transcriptional on and off states and are essential for cells to respond to environmental cues with high fidelity. However, conventional switches, which rely on direct repressor-DNA binding, are extremely noise-sensitive, leading to unintended changes in gene expression. Here, through model simulations and analysis, we discovered that an alternative design combining three indirect transcriptional repression mechanisms, sequestration, blocking, and displacement, can generate a noise-resilient ultrasensitive switch. Although sequestration alone can generate an ultrasensitive switch, it remains sensitive to noise because the unintended transcriptional state induced by noise persists for long periods. However, by jointly utilizing blocking and displacement, these noise-induced transitions can be rapidly restored to the original transcriptional state. Because this transcriptional switch is effective in noisy cellular contexts, it goes beyond previous synthetic transcriptional switches, making it particularly valuable for robust synthetic system design. Our findings also provide insights into the evolution of robust ultrasensitive switches in cells. Specifically, the concurrent use of seemingly redundant indirect repression mechanisms in diverse biological systems appears to be a strategy to achieve noise-resilience of ultrasensitive switches. [ABSTRACT FROM AUTHOR]

Accurately detecting distant evolutionary relationships between proteins remains an ongoing challenge in bioinformatics. Search methods based on primary sequence struggle to accurately detect homology between sequences with less than 20% amino acid identity. Profile- and structure-based strategies extend sensitive search capabilities into this twilight zone of sequence similarity but require slow pre-processing steps. Recently, whole-protein and positional embeddings from deep neural networks have shown promise for providing sensitive sequence comparison and annotation at long evolutionary distances. Embeddings are generally faster to compute than profiles and predicted structures but still suffer several drawbacks related to the ability of whole-protein embeddings to discriminate domain-level homology, and the database size and search speed of methods using positional embeddings. In this work, we show that low-dimensionality positional embeddings can be used directly in speed-optimized local search algorithms. As a proof of concept, we use the ESM2 3B model to convert primary sequences directly into the 3D interaction (3Di) alphabet or amino acid profiles and use these embeddings as input to the highly optimized Foldseek, HMMER3, and HH-suite search algorithms. Our results suggest that positional embeddings as small as a single byte can provide sufficient information for dramatically improved sensitivity over amino acid sequence searches without sacrificing search speed. [ABSTRACT FROM AUTHOR]