Spoken language identification has gained a lot of importance in the last decade with the advent of text-to-speech systems and intelligent assistants. Language Identification (LID) task revolves around the automated recognition of the language spoken in a given speech utterance. The traditionally used systems in this scenario are GMMs and SVMs. However, these systems also carry disadvantages like limitations in speed and size, discrete data handling, high algorithmic complexity and extensive memory requirements in large scale tasks. Deep Learning, a subset of machine learning, is pushing the boundaries of technological advancements in areas such as spoken language identification. It has already been proven that deep neural networks can be used to efficiently train the speech recognition system. A convolutional neural network refers to a feedforward artificial neural network containing multiple layers of hidden units positioned between its inputs and outputs. In this research work, we have generated melspectrogram images from a collection of speech utterances obtained from the open source website Voxforge. These melspectrograms are used to train the Convolutional Neural Network. The performance of the proposed spoken language identification using three languages namely French, English and Spanish are significantly better than the traditional machine learning approach.

In recent years, a significant portion of the content on various platforms on the internet has been found to be offensive or abusive. Abusive comment detection can go a long way in preventing internet users from facing the adverse effects of coming in contact with abusive language. This problem is particularly challenging when the comments are found in low-resource languages like Tamil or Tamil-English code-mixed text. So far, there has not been any substantial work on abusive comment detection using imbalanced datasets. Furthermore, significant work has not been performed, especially for Tamil code-mixed data, that involves analysing the dataset for classification and accordingly creating a custom vocabulary for preprocessing. This paper proposes a novel approach to classify abusive comments from an imbalanced dataset using a customised training vocabulary and a combination of statistical feature selection with language-agnostic feature selection while making use of explainable AI for feature refinement. Our model achieved an accuracy of 74% and a macro F1-score of 0.46.

This paper explores the transformative impact of the IoT on device connectivity, emphasizing the critical role of Access Control (AC) in securing IoT and ensuring appropriate resource access. The study conducts a comprehensive survey, addressing challenges, requirements, and approaches in designing effective IoT AC, considering device diversity, resource constraints, and energy consumption complexities. The adaptation of traditional AC methods for IoT is examined, highlighting complications arising from device heterogeneity and advocating for suitable AC solutions aligned with global IoT policies. The focus shifts to two emerging technologies, Blockchain (BC) and Machine Learning (ML), as potential solutions for IoT AC challenges. BC’s decentralization, security, and transparency are explored, emphasizing their role in mitigating AC issues. ML techniques are investigated for real-time identification of suspicious activities, enhancing the adaptability of AC mechanisms. Integrating BC and ML in IoT AC is a key research question, with insights suggesting enhanced security, transparency, and efficiency. The survey employs a systematic review approach, utilizing NVIVO software and the CASP tool for data coding to ensure the validity of the information. The conclusion summarizes key findings, addressing how BC and ML can mitigate AC challenges in IoT. BC’s decentralized and tamper-resistant nature and ML’s dynamic learning capabilities contribute to enhanced security. Alternative technologies for IoT AC are explored, including smart contracts, attribute-based AC, and certificate-based schemes. The integration of BC and ML is identified as a means to deliver trusted IoT services, achieve fine-grained AC, and enable efficient data transfer methods.

The proposed research work addresses the challenge of developing a near real-time Tamil dialect-based transcription system. The primary aim is to improve automatic speech recognition (ASR) for Tamil dialects, focusing on minimizing latency and enhancing accuracy and providing an open-source solution to tackle the problem of recognizing accented speech. In this regard, we introduce a novel Tamil dialect-based speech and text corpus to fine-tune a whisper model, specifically the whisper small architecture, to generate accurate transcripts. To overcome resource constraints, we employed a parameter-efficient fine-tuning technique called low ranking adapters. The fine-tuned models achieved a 61

Robust fake speech detection systems are crucial in an era where audio recordings can be easily altered or developed due to advancements in technology. The potential impact of this technology could be devastating due to its susceptibility to misuse. It can lead to political and economic instability, severe financial repercussions, the spread of misinformation, defamation, and security breaches in ASV systems and facilitate theft and fraud. Further, existing detection systems face a key issue; most state-of-the-art systems only provide results on the corpus they are trained on but fail in the domain variability scenario. The goal of this study is to enable a system to generalize across various domains, enhancing its reliability in real-world scenarios and thereby solidifying the authenticity of speech, making speech-based systems more dependable. All three audio spoofing scenarios are considered, logical access-based (LA) attacks including TTS and voice conversion (VC) techniques, spoofing attacks generated in real physical space, namely physical access (PA) attacks (replay, mimicry) and advanced deepfake technologies. The proposed system put forth in this study achieves commendable performance on three diverse test datasets. Notably, EER of 0.286, 0.337 and 0.371 is achieved on the In-The-Wild dataset with the proposed system implemented on ResNet, ECANet and SENet.

The process of turning spoken language into written text is called automatic speech recognition (ASR). It is used in many o settings. Automatic speech recognition becomes a crucial tool when daily life is digitized. It is well known that it considerably improves the lives of the elderly and people with disabilities. Minor dysarthria, or slurred speech, is common in elderly people and those who are physically or mentally challenged, which leads to erroneous transcription of the data. In this study, we suggested creating a Tamil-language automatic voice recognition system for the elderly. The ASR system must be trained using elderly people's speech utterances in order to increase its performance when processing elderly people's speech. There is no Tamil speech corpus made up of elderly speakers. We recorded elderly and transgender individuals speaking Tamil on the spot. These statements were gathered from people speaking in open spaces including markets, hospitals, and vegetable shops. The speech corpus contains remarks from men, women, and transgender people. In this research, an attention-based, end-to-end paradigm is used to construct an ASR system. The proposed system includes two key steps: creating an audio model and a linguistic model. Recurrent neural network architecture was used in the construction of the language model. The attention-based encoder–decoder architecture was used to construct the acoustic model. The encoder model utilized a convolution network with a recurrent network, and the decoder model utilized an attention-based gated recurrent unit. Word error rate (WER) is used to assess how well the suggested ASR system performs when used on geriatric speaking utterances. The outcomes are compared to several transformer models that have already been trained. By pretraining a single model using the raw waveform of speech in various languages, the pre-trained XLSR models develop cross-lingual speech representations. The Common Voice Tamil voice corpus is used to fine tune the pre-trained models. According to the experiments, the suggested attention-based, end-to-end model performs noticeably better than transformer models that have already been trained.

Tamil is one of the oldest languages that possesses a vast knowledge of literature and grammar. Dialect refers to the language that the majority of people in an area speak. Speech recognition technology has become increasingly common in modern electronics. Dialects of a given language play a crucial part in this. Because there are many similarities among dialects of language, discriminating between them is a critical component of dialect identification systems. The dialect of a language is not a distinctive characteristic because nativity influences it. The formation of dialects can be for different reasons. Automatic speech recognition (ASR) will be impacted by dialect identification since speakers who are accustomed to a variety of dialects will pronounce the same words differently. There will be dialect information in speech at various segmental levels. The creation of an appropriate technique for the precise detection of dialect will aid in applications and support, such as the speech recognition technology used in practically all electronic gadgets, are being improved. Due to the greater resemblance across dialects of the same language, dialect identification is regarded as a more challenging topic than language recognition. In this work, identification of dialects of the Tamil language is explored. For speech processing research, until today, there is no standard speech database created for Tamil dialects. The proposed system aims to develop a speech corpus that can be used for recognition of Tamil dialects. Three dialects such as southern, northern, and western dialects are considered for this study. Mel-frequency cepstral coefficients (MFCC) are extracted from the speech utterances. Gaussian mixture models (GMM) are trained using the extracted features of the three dialects. Mel-frequency cepstral coefficient features of the test utterances are given to the three GMM models to recognize the dialect. The performances of the dialect identification system for southern and northern dialects are better than western dialects. The number of mixture components for the GMM is varied and the performance is analyzed.

The objective of this work is to classify the mind mapping decisions “like”, “dislike” and “neutral” in Epilepsy patients by applying the concepts of rough sets. An effective rough set-based classification of mental status in epilepsy patients has been computed using the features such as meditation, familiarity, theta, attention, appreciation, beta, mental effort, delta, alpha and gamma. The significance of features is considered as conditional attributes and the expected mood is represented as decision attributes. To analyze the impact of the features, the cardinality and rough set-based approximation are computed. Grey Relational Analysis (GRA) algorithm is applied for classification of patient decision is either like or dislike or neutral. The experimental results on classification of mind mapping of epilepsy patients using rough set-based approximation yields 95% accuracy.

Artificial Intelligence (AI) has led to advancements in multiple fields of research, and music has always been a field of high interest. Music is an important part of life and various studies have shown the link between better living and listening to music. From completing melodies to composing music from scratch, there are many applications of AI in this domain. This paper aims to analyse one such application of using AI for music generation, specifically for instrumental accompaniment. Instrumental accompaniment is essentially the instrumental music that is composed to support or complement a melody. Creating instrumental accompaniment for music generally requires extensive musical knowledge or forming a band together with skilled instrumentalists. Build Your Own Band (BYOB) attempts to simplify this process with the help of AI. In this research work, three transformer models are employed for training various accompanying instruments. Here, the proposed transformer model accepts a melody line as input and produces an accompanying track with instruments like bass instruments, the guitar and string instruments. One of the main challenges was to make sure that these instruments produce a cohesive sound.

As the world's population grows, there are more vehicles on the road every day, which leads to an increase in heavy traffic. Traffic monitoring is essential for preventing accidents. To detect reckless drivers and other traffic infractions, a model that can track, identify, and categorize vehicles is needed. The task of counting the number of vehicles is crucial in traffic situations because it allows the authorities to prevent accidents and traffic jams caused by heavy traffic. The approach outlined in the study uses the image processing methods YOLO and OpenCV to count the number of vehicles, classify them, and identify them. By processing the images from the input video given to OpenCV, a software library, the objects are detected and identified. In comparison to other object detection algorithms, the real-time object detection algorithm YOLO is both quicker and more accurate. The accuracy and efficiency of vehicle detection and classification have been greatly enhanced by convolutional neural networks and other machine learning algorithms, enabling real-time analysis of enormous amounts of data. With the help of this technology, driving safety will be increased, traffic flow will be optimized, and autonomous driving will be made possible.

Internet of Things (IoT) devices are connected network smart elements, sensors, and software capabilities that allow for data acquisition and transact via the internet. These days, it's very difficult to secure the devices used in the IoT applications. This article talks about concise review of the security challenges in IoT devices, kinds of attacks, IoT security software tools and available directives for IoT system design. The main contribution is the change in emphasis from architecture to devices and the suggestion of technology advancements needed for IoT device security. This investigation also studies about standards and certifying procedures currently in vogue and suggests a route for further research. It also discusses the current testing techniques and technological advancements that will affect device certification in the future.

Communication has been an essential part of human life. Different languages are present around the world for communication. Still, people who have lost their hearing and speaking ability by accidents and genetic birth often face difficulty in communication. Auditory-impaired people have found sign language helpful in communicating with others. Hearing-impaired people must always have a personal interpreter available for translations whenever they need to communicate. People with disability of hearing impairment find it challenging to interact with others on social media and the internet to form new relationships on their own. An open-source video-conferencing application that can translate sign language is quite helpful for the hearing impaired. Sign language recognition (SLR) has drawn a lot of attention as a way to close the enormous communication gap. However, sign language is far more complex and unpredictable when compared to other activities, making it challenging for reliable recognition. The Speech-to-Text API enables speech-impaired people who can read to comprehend others. The Sign Language Translation Application (SLTA) allows them to communicate by translating their sign language into text that others can understand. The proposed method uses python, the MediaPipe Framework for gesture data extraction, and the Deep Gesture Recognition (DGR) Model to identify the sign motion in real-time. The proposed method achieves the highest accuracy of 98.81% using a neural network comprised of Long-Short Term Memory units for sequence identification.

Identifying sarcasm in code-mixed languages presents a pivotal challenge within the domain of natural language processing, given the widespread use of multilingual and multicultural communication on social media platforms. This paper presents findings of the Sarcasm Identification shared task hosted as part of DravidianCodeMix@FIRE-2023. The principal objective of this task was to detect instances of sarcasm within a dataset of code-mixed comments in Tamil-English and Malayalam-English, collected from social media platforms. A total of 11 teams participated in this collaborative effort. This analysis provides a comprehensive assessment of the diverse models employed by the participating teams and the system’s performance was measured in terms of the macro-F1 score.

The necessity to ensure that Internet of Things (IoT) networks are secure is one of the biggest issues that has arisen as a result of the growing demand for technology that uses the IoT. Considering how many gadgets are linked to the internet, safeguarding their networks is a growing worry. Due to the IoT's network's complexity and resource constraints, traditional intrusion detection systems encounter a number of problems. The main objectives of this project are to design, develop, and evaluate a hybrid level placement method for an IDS based on multi- agent systems, BC technology (Block-Chain), and DL algorithms (Deep Learning). The breakdown of data administration, data collection, analysis, and reaction into its component parts reveals the overall system design. The National Security Laboratory's knowledge discovery and data mining dataset is used to test the system as part of the validation procedure. These results demonstrate how deep learning algorithms are effective at identifying risks at the network and transport levels. The experiment shows that deep learning techniques function well when used to find intrusions in a network environment for the Internet of Things.

Maximum user systems on 5G networks will now not be consumer phones or computers, but IoT device. Via 2021, there might be about 30 billion such devices. The quantity of attacks on the IoT is growing. Device protection is terrible and malware distribution is without problems scalable. Protection has ended up the primary challenge in many telecommunications industries these days as risks may have high outcomes. especially, because the center and enable technologies might be related to the 5G network, the confidential information will pass at all layers in destiny Wi-Fi structures. Even with modern-day 4G networks, now not each operator succeeds in securely configuring the center network and protecting it from all angles. As SDN and NFV are carried out for network cutting in 5G, the administration will become even extra difficult. Flexibility in 5G networks comes at the fee of multiplied complexity and high bandwidth communication settings to monitor. 5G will offer broadband access anywhere, entertain better person mobility, and permit connectivity of a large number of devices in an ultra- reliable and low-priced manner. Furthermore, we present protection solutions to those demanding situations and future instructions for secure 5G systems.

We present an overview of sentiment analysis and homophobia detection of YouTube comments in code-mixed Dravidian languages in this paper. We provide the details of this task and the submitted systems for the tasks. We introduce two studies: task A for detecting sentiment analysis and task B on homophobia detection, which is organized by the FIRE 2022. A total of 95 participants registered for the shared task, 13 teams finally submitted their results for task-A a, and 10 teams submitted their results for task B. The teams explored tasks A and B using traditional machine learning and deep learning models. Most of the benchmark systems have been analyzed by participants capable of handling code-mixed scenarios in Dravidian languages.