Voice Activity Detection Research Articles

Fine-grained hate speech detection in Arabic using transformer-based models

With the proliferation of social media platforms, characterized by features such as anonymity, user-friendly access, and the facilitation of online community building and discourse, the matter of detecting and monitoring hate speech has emerged as an increasingly formidable challenge for society, individuals, and researchers. Despite the crucial importance of hate speech detection task, the majority of work in this field has been conducted in English, with insufficient focus on other languages, particularly Arabic. Furthermore, most existing studies on Arabic hate speech detection have addressed this task as a binary classification problem, which is unreliable. Therefore, the aim of this study is to provide an enhanced model for detecting fine-grained hate speech in Arabic. To this end, three transformer-based models were evaluated to generate contextualized word embeddings from input sequence. Additionally, these models were combined with a bidirectional gated recurrent unit (BiGRU) layer to further improve the extracted semantic and context features. The experiments were conducted on an Arabic reference dataset provided by the open-source Arabic corpora and processing tools (OSACT-5) shared task. A comparative analysis indicates the efficiency of the proposed model over the baseline and related work models by achieving a macro F1-score of 61.68%.

Domain adaptation-based method for improving generalization of hate speech detection models

Domain adaptation-based method for improving generalization of hate speech detection models

Automatic hate speech detection in audio using machine learning algorithms

Automatic hate speech detection in audio using machine learning algorithms

Robustness of models addressing Information Disorder: A comprehensive review and benchmarking study

Machine learning and deep learning models are increasingly susceptible to adversarial attacks, particularly in critical areas like cybersecurity and Information Disorder. This study provides a comprehensive evaluation of model Robustness against such attacks across key tasks well-assessed in Information Disorder literature: Toxic Speech Detection, Sentiment Analysis, Propaganda Detection, and Hate Speech Detection. Rigorous experiments conducted across 13 models and 12 diverse datasets highlight significant vulnerabilities. The methodological framework implements adversarial attacks that strategically manipulates model inputs based on keyword significance, identified using the LIME method, an advanced explainable AI technique. The evaluation measures Robustness primarily through accuracy of the models and attack success rates. The experiments reveal that current models display inconsistent resistance to adversarial manipulations, underscoring an urgent need for developing more sophisticated defensive strategies. The study sheds light on the critical weaknesses in existing models and charts a course for future research to fortify AI resilience against evolving cyber threats. The findings advocate for a paradigm shift in model training and development to prioritize adversarial Robustness, ensuring that AI systems are equipped to handle real-world adversarial scenarios effectively.

Hate Speech Detection Using Deep Learning

Abstract: Hate speech is a form of verbal abuse that targetsindividuals depending on their race, religion, ethnicity, gender, sexual orientation, or other personal characteristics. It can be spoken, written, or displayed, and can be found in a variety of contexts, including online, in person, and in the media. Hate Speech has the potential to have a devastating impact on victims, causing emotional distress, social isolation, and even physical harm. Recognizing hate speech is a challenging task, especially in multilingual contexts. This is because hate speech can be expressed in many ways, and can be difficult todistinguish from other forms of speech. However, Artificial Intelligence (AI) has advanced recently, and have made it possible to develop effective hate speech detection systems. Hate speech detection is a challenging task, especially in multilingual contexts. This survey paper reviews the recent advances in hate speech detection using BERT and CNN models. We explored the various approaches that have been proposed, the challengesfaced, and the paths that research will take in the future

A comprehensive review on Arabic offensive language and hate speech detection on social media: methods, challenges and solutions

A comprehensive review on Arabic offensive language and hate speech detection on social media: methods, challenges and solutions

AUTOMATED HATE SPEECH DETECTION USING RECURRENT NEURAL NETWORK

The spread of online entertainment and information sharing has indeed brought significant advances to humanity. However, it has also created various challenges, such as the spread of derogatory content through means such as voicemail. Addressing this emerging problem requires continued consideration, especially in virtual environments, where innovative approaches combining design methods and artificial intelligence are used to identify and filter such content from various data sets. While efforts are being made to combat this problem, there is still a gap in design strategies for that purpose and a comprehensive overview of artificial intelligence algorithms. This includes the need for transparent evaluation criteria and publicly available datasets to benchmark performance and promote collaboration within the research community. Keywords — Deep Learning, Recurrent Neural Network(RNN), Hate Speech Detection.

Highly Sensitive Biomimetic Crack Pressure Sensor with Selective Frequency Response.

High-sensitivity sensors in practical applications face the issue of environmental noise interference, requiring additional noise reduction circuits or filtering algorithms to improve the signal-to-noise ratio (SNR). To address this issue, this study proposes a biomimetic crack pressure sensor with selective frequency response based on hydrogel dampers. The core of this research is to construct a biomimetic crack pressure sensor with selective frequency response using the high-pass filtering characteristics of gelatin-chitosan hydrogels. This design, inspired by the slit sensilla and stratum corneum structure of spider legs, delves into the material properties and principles of hydrogel dampers, exploring their application in biomimetic crack pressure sensors, including parameter selection, structural design, and performance optimization. By delving into the nuanced characteristics and working principles of hydrogel dampers, their integration in biomimetic crack pressure sensors is examined, focusing on aspects like parameter selection, structural engineering, and performance enhancement to selectively sieve out low-frequency noise and transmit target vibrational signals. Experimental results demonstrate that this innovative sensor, while suppressing low-frequency vibration signals, can selectively detect high-frequency signals with high sensitivity. At different vibration frequencies, the relative change in resistance exceeds 200%, and the sensor sensitivity is 7 × 104 kPa-1. Furthermore, this sensor was applied to human voice detection, and the corresponding results verified its frequency-selective performance evidently. This study not only proposes a new design for pressure sensors but also offers fresh insights into the application of biomimetic crack pressure sensors in intricate environments.

MetaHate: A Dataset for Unifying Efforts on Hate Speech Detection

MetaHate: A Dataset for Unifying Efforts on Hate Speech Detection

Exploring the Effectiveness of the Phase Features on Double Compressed AMR Speech Detection

Determining whether an audio signal is single compressed (SC) or double compressed (DC) is a crucial task in audio forensics, as it is closely linked to the integrity of the recording. In this paper, we propose the utilization of phase spectrum-based features for detecting DC narrowband and wideband adaptive multi-rate (AMR-NB and AMR-WB) speech. To the best of our knowledge, phase spectrum features have not been previously explored for DC audio detection. In addition to introducing phase spectrum features, we propose a novel parallel LSTM system that simultaneously learns the most representative features from both the magnitude and phase spectrum of the speech signal and integrates both sets of information to further enhance its performance. Analyses demonstrate significant differences between the phase spectra of SC and DC speech signals, suggesting their potential as representative features for DC AMR speech detection. The proposed phase spectrum features are found to perform as well as magnitude spectrum features for the AMR-NB codec, while outperforming the magnitude spectrum in detecting AMR-WB speech. The proposed phase spectrum features yield 8% performance improvement in terms of true positive rate over the magnitude spectrogram features. The proposed parallel LSTM system further improves DC AMR-WB speech detection.

Automatic detection of hate speech in code-mixed Indian languages in twitter social media interaction using DConvBLSTM-MuRIL ensemble method

Automatic detection of hate speech in code-mixed Indian languages in twitter social media interaction using DConvBLSTM-MuRIL ensemble method

StereoHate: Toward identifying stereotypical bias and target group in hate speech detection

Abstract Though social media helps spread knowledge more effectively, it also stimulates the propagation of online abuse and harassment, including hate speech. It is crucial to prevent hate speech since it may have serious adverse effects on both society and individuals. Therefore, it is not only important for models to detect these speeches but to also output explanations of why a given text is toxic. While plenty of research is going on to detect online hate speech in English, there is very little research on low-resource languages like Hindi and the explainability aspect of hate speech. Recent laws like the “right to explanations” of the General Data Protection Regulation have spurred research in developing interpretable models rather than only focusing on performance. Motivated by this, we create the first interpretable benchmark hate speech corpus hate speech explanation (HHES) in the Hindi language, where each hate post has its stereotypical bias and target group category. Providing descriptions of internal stereotypical bias as an explanation of hate posts makes a hate speech detection model more trustworthy. Current work proposes a commonsense-aware unified generative framework, CGenEx, by reframing the multitask problem as a text-to-text generation task. The novelty of this framework is it can solve two different categories of tasks (generation and classification) simultaneously. We establish the efficacy of our proposed model (CGenEx-fuse) on various evaluation metrics over other baselines when applied to the Hindi HHES dataset. Disclaimer The article contains profanity, an inevitable situation for the nature of the work involved. These in no way reflect the opinion of authors.

Faked speech detection with zero prior knowledge

Faked speech detection with zero prior knowledge

Deep Learning Approaches for Hate Speech Detection

The main goals of this project are to develop reliable hate speech identification models that can recognize derogatory terminology, prejudiced attitudes, and damaging stereotypes on a variety of internet platforms. To help the models train and generalize efficiently, the study focuses on using big datasets with a variety of content types, including both hate speech and non-hate speech. The findings of this study suggest that machine learning has the potential to mitigate the negative consequences of hate speech by using automated filtering and flagging tools. The study also highlights the need for continued research and development to improve the accuracy, uniformity, and transparency of hate speech detection systems, and ultimately to foster a safer online environment to encourage all people.

CHUNAV: Analyzing Hindi Hate Speech and Targeted Groups in Indian Election Discourse

In the ever-evolving landscape of online discourse and political dialogue, the rise of hate speech poses a significant challenge to maintaining a respectful and inclusive digital environment. The context becomes particularly complex when considering the Hindi language—a low-resource language with limited available data. To address this pressing concern, we introduce the CHUNAV dataset—a collection of 11,457 Hindi tweets gathered during assembly elections in various states. CHUNAV is purpose-built for hate speech categorization and the identification of target groups. The dataset is a valuable resource for exploring hate speech within the distinctive socio-political context of Indian elections. The tweets within CHUNAV have been meticulously categorized into “Hate” and “Non-Hate” labels, and further subdivided to pinpoint the specific targets of hate speech, including “Individual”, “Organization”, and “Community” labels (as shown in Figure 1). Furthermore, this paper presents multiple benchmark models for hate speech detection, along with an innovative ensemble and oversampling-based method. The paper also delves into the results of topic modeling, all aimed at effectively addressing hate speech and target identification in the Hindi language. This contribution seeks to advance the field of hate speech analysis and foster a safer and more inclusive online space within the distinctive realm of Indian Assembly Elections.

Transfer effects of recursive self-feedback on connected speech production in patients with chronic nonfluent aphasia: Preliminary results

ABSTRACT Background Persons with nonfluent aphasia (PWNA) have deficits with online feedback for speech error detection and correction which have debilitating effects on language relearning in learned and novel contexts. Previous studies have shown that aphasia treatments that combine PWNA’s offline self-feedback and external feedback from speech-language pathologists (SLPs) improve language production in learned and novel contexts. However, it remains unclear whether PWNA can use their offline self-feedback alone during treatments that generalizes to improvement in connected speech production. Here, we examine generalization effects of recursive self-feedback on PWNA’s connected speech production. Recursive self-feedback utilizes offline self-feedback loops alone to help PWNA optimize their error monitoring and correction abilities for language production in learned and novel contexts. Method A crossover design was used to investigate generalization effects of two treatments on connected speech production in four PWNA. These treatments comprised script production with recursive self-feedback and script production with external feedback. Employing a mobile health approach, treatments were remotely administered to participants within the comfort of their homes. Practice of both treatment blocks occurred intensively over two hours per day, five days per week, spanning two to three weeks. Each treatment block featured pretreatment and posttreatment assessment phases, with the sequencing of treatment blocks evenly counterbalanced across all participants. Following the initial treatment block, a two-week period devoid of treatment preceded the commencement of the second treatment block. Untrained narrative elicitation tasks were utilized to prompt connected speech production samples from all participants during the assessment phases of each treatment block. Effect sizes per participant were computed using within-case standard mean difference analysis. Results PWNA exhibited variability in performance, showing improvement in microlinguistic measures of connected speech following both script production with recursive self-feedback and script production with external feedback. Particularly, script production with recursive self-feedback demonstrated marginally superior enhancement in connected speech measures compared to its counterpart. Conclusion The study provides preliminary evidence for generalization effects of recursive self-feedback training and external feedback training on PWNA’s connected speech production. Further research is required to corroborate the findings and to implement recursive self-feedback procedures optimized specifically for spontaneous speech.

Bridging the gap in online hate speech detection: A comparative analysis of BERT and traditional models for homophobic content identification on X/Twitter

Bridging the gap in online hate speech detection: A comparative analysis of BERT and traditional models for homophobic content identification on X/Twitter

War of Words: Harnessing the Potential of Large Language Models and Retrieval Augmented Generation to Classify, Counter and Diffuse Hate Speech

This paper explores the emergence of divergent narratives in the wake of the Russian-Ukraine war, which began on February 24, 2022, and the innovative application of AI language models, specifically RetrievalAugmented Generation (RAG) and instruction-based large language models (LLMs), in countering hateful speech on social media. We design a pipeline to automatically discover and then respond to hateful content trending on social media platforms. Monitoring via traditional topic/narrative modeling often focuses on lowlevel content, which is difficult to interpret. In addition, workflows for prioritization and response generation are often highly manual. We utilize several large language models (LLMs) throughout our pipeline to detect and summarize topics, to determine whether tweets contain hate speech and to generate counter narratives. We test our approach on Ukraine Bio-Lab Tweet Corpus of 500k Tweets and evaluate the counter-narrative generation performance across several dimensions: relevance, grammaticality, factuality, and diversity. Our approach outperforms existing state of the art algorithms for hate speech detection and promising counter-narrative generation performance scores across our metrics reflect effectiveness of our pipeline in addressing hateful social media posts

Impact of Data Augmentation on the Hate Speech Detection in Portuguese Language

Online communities allow users to establish a web presence, manage their identities, and stay connected with others. The internet has facilitated global outreach with just a click on the World Wide Web. However, the current landscape of online social media platforms are marred by various issues, with hate speech prominently taking center stage. Hate speech is characterized by hostile and malicious language driven by prejudice, targeting individuals or groups based on their innate, natural, or perceived characteristics. Detecting such speech is crucial for maintaining a safe online environment. This study examines the impact of dataset regularization techniques on the performance of BERTimbau-based models when applied to four Portuguese hate speech datasets: Fortuna et al. (2019), OFFCOMBR-2, ToLD-BR, and Hate-BR. Four Data Augmentation techniques are evaluated: Oversampling, Undersampling, Text Augmentation, and Synonym Replacement. Our experiments revealed that, apart from the Fortuna et al. (2019) dataset, the Data Augmentation techniques did not significantly enhance the performance of hate speech detection tasks.

End-to-end integration of speech separation and voice activity detection for low-latency diarization of telephone conversations

Recent works show that speech separation guided diarization (SSGD) is an increasingly promising direction, mainly thanks to the recent progress in speech separation. It performs diarization by first separating the speakers and then applying voice activity detection (VAD) on each separated stream. In this work we conduct an in-depth study of SSGD in the conversational telephone speech (CTS) domain, focusing mainly on low-latency streaming diarization applications. We consider three state-of-the-art speech separation (SSep) algorithms and study their performance both in online and offline scenarios, considering non-causal and causal implementations as well as continuous SSep (CSS) windowed inference. We compare different SSGD algorithms on two widely used CTS datasets: CALLHOME and Fisher Corpus (Part 1 and 2) and evaluate both separation and diarization performance. To improve performance, a novel, causal and computationally efficient leakage removal algorithm is proposed, which significantly decreases false alarms. We also explore, for the first time, fully end-to-end SSGD integration between SSep and VAD modules. Crucially, this enables fine-tuning on real-world data for which oracle speakers sources are not available. In particular, our best model achieves 8.8% DER on CALLHOME, which outperforms the current state-of-the-art end-to-end neural diarization model, despite being trained on an order of magnitude less data and having significantly lower latency, i.e., 0.1 vs. 1 s. Finally, we also show that the separated signals can be readily used also for automatic speech recognition, reaching performance close to using oracle sources in some configurations.