Search results for: speech recognition performance

Authors: Gunjita Shami, Noopur Anand

Abstract:

The pilot run of 50 days was undertaken to test employability potential of people with visual and hearing & speech impairment. Various roles in an apparel manufacturing set up like spreading of fabric for cutting, folding, sealing and labeling cartons, pasting size barcode stickers on packed garments, removing tickets from the garments in the finishing stage were studied. Their performance was quantified basis timesheets for all the days and improvement per day was quantified. Their final day output was compared to that of the able-bodied worker. For example in the carton making activity on day one visually impaired worker was making one box every three minutes which improved to four boxes per minute on day 28 displaying 91.6% improvement compared or an improvement of 3.6% per day which was comparable to the able-bodied seasoned workers, who were making 5 boxes per minute. The performance of persons with hearing and speech impairment in the finishing department was 10% higher than that of able-bodied seasoned workers in the same process. Overall in all the activities the differently abled showed day to day improvement of 65% while able bodied displayed improvement of 52%. On the first day performance of able-bodied worker was 75% better than that of differently abled while on the 50th day it was only 20% better. Therefore the performance of persons with disabilities was found comparable to the able bodied person. The results, though on a small scale, showed a big promise of employment of persons with disability in the apparel industry. Armed with the promising result a full-scale study has been undertaken to identify the roles suitable for certain kind of disability in apparel production, work-aids required to assist the differently abled to improve performance and measures to be undertaken to make production floor 'friendlier' for them. The results have been discussed in this paper which opens doors for integrating differently abled into the world projected and assumed for only able-bodied.

Keywords: apparel sector, differently abled, employability, performance, work-aid

Procedia PDF Downloads 123

Authors: Yuan-Hsiang Chang, Zhong-Xian Peng, Li-Der Jeng

Abstract:

Music has always been an integral part of human’s daily lives. But, for the most people, reading musical score and turning it into melody is not easy. This study aims to develop an Automatic music score recognition system using digital image processing, which can be used to read and analyze musical score images automatically. The technical approaches included: (1) staff region segmentation; (2) image preprocessing; (3) note recognition; and (4) accidental and rest recognition. Digital image processing techniques (e.g., horizontal /vertical projections, connected component labeling, morphological processing, template matching, etc.) were applied according to musical notes, accidents, and rests in staff notations. Preliminary results showed that our system could achieve detection and recognition rates of 96.3% and 91.7%, respectively. In conclusion, we presented an effective automated musical score recognition system that could be integrated in a system with a media player to play music/songs given input images of musical score. Ultimately, this system could also be incorporated in applications for mobile devices as a learning tool, such that a music player could learn to play music/songs.

Keywords: connected component labeling, image processing, morphological processing, optical musical recognition

Procedia PDF Downloads 393

Authors: Sonali Banrjee, Swarup Kumar Mitra, Aritra Acharyya

Abstract:

A piano note recognition method has been proposed by the authors in this paper. The authors have used a comprehensive method for onset detection of each note present in a piano piece followed by segmented short-time Fourier transform (STFT) for the identification of piano notes. The performance evaluation of the proposed method has been carried out in different harsh noisy environments by adding different levels of additive white Gaussian noise (AWGN) having different signal-to-noise ratio (SNR) in the original signal and evaluating the note detection error rate (NDER) of different piano pieces consisting of different number of notes at different SNR levels. The NDER is found to be remained within 15% for all piano pieces under consideration when the SNR is kept above 8 dB.

Keywords: AWGN, onset detection, piano note, STFT

Procedia PDF Downloads 142

Authors: Shanxiong Chen, Xu Han, Xiaolong Wang, Hui Ma

Abstract:

Yi is an ethnic group mainly living in mainland China, with its own spoken and written language systems, after development of thousands of years. Ancient Yi is one of the six ancient languages in the world, which keeps a record of the history of the Yi people and offers documents valuable for research into human civilization. Recognition of the characters in ancient Yi helps to transform the documents into an electronic form, making their storage and spreading convenient. Due to historical and regional limitations, research on recognition of ancient characters is still inadequate. Thus, deep learning technology was applied to the recognition of such characters. Five models were developed on the basis of the four-layer convolutional neural network (CNN). Alpha-Beta divergence was taken as a penalty term to re-encode output neurons of the five models. Two fully connected layers fulfilled the compression of the features. Finally, at the softmax layer, the orthographic features of ancient Yi characters were re-evaluated, their probability distributions were obtained, and characters with features of the highest probability were recognized. Tests conducted show that the method has achieved higher precision compared with the traditional CNN model for handwriting recognition of the ancient Yi.

Keywords: recognition, CNN, Yi character, divergence

Procedia PDF Downloads 141

Authors: Nayra Martin-Key, Erich Graf, Wendy Adams, Graeme Fairchild

Abstract:

Children and adolescents with Conduct Disorder (CD) have been shown to demonstrate impairments in emotion recognition, but it is currently unclear whether this deficit is related to specific emotions or whether it represents a global deficit in emotion recognition. An emotion recognition task with concurrent eye-tracking was employed to further explore this relationship in a sample of male and female adolescents with CD. Participants made emotion categorization judgements for presented dynamic and morphed static facial expressions. The results demonstrated that males with CD, and to a lesser extent, females with CD, displayed impaired facial expression recognition in general, whereas callous-unemotional (CU) traits were linked to specific problems in sadness recognition in females with CD. A region-of-interest analysis of the eye-tracking data indicated that males with CD exhibited reduced fixation times for the eye-region of the face compared to typically-developing (TD) females, but not TD males. Females with CD did not show reduced fixation to the eye-region of the face relative to TD females. In addition, CU traits did not influence CD subjects’ attention to the eye-region of the face. These findings suggest that the emotion recognition deficits found in CD males, the worst performing group in the behavioural tasks, are partly driven by reduced attention to the eyes.

Keywords: attention, callous-unemotional traits, conduct disorder, emotion recognition, eye-region, eye-tracking, sex differences

Procedia PDF Downloads 285

Authors: Marcio Leal, Marta Villamil

Abstract:

Computacional recognition of sign languages aims to allow a greater social and digital inclusion of deaf people through interpretation of their language by computer. This article presents a model of recognition of two of global parameters from sign languages; hand configurations and hand movements. Hand motion is captured through an infrared technology and its joints are built into a virtual three-dimensional space. A Multilayer Perceptron Neural Network (MLP) was used to classify hand configurations and Dynamic Time Warping (DWT) recognizes hand motion. Beyond of the method of sign recognition, we provide a dataset of hand configurations and motion capture built with help of fluent professionals in sign languages. Despite this technology can be used to translate any sign from any signs dictionary, Brazilian Sign Language (Libras) was used as case study. Finally, the model presented in this paper achieved a recognition rate of 80.4%.

Keywords: artificial neural network, computer vision, dynamic time warping, infrared, sign language recognition

Procedia PDF Downloads 189

Authors: Rajpal Kaur, Pooja Choudhary

Abstract:

Biometrics, which refers to identifying an individual based on his or her physiological or behavioral characteristics, has the capabilities to the reliably distinguish between an authorized person and an imposter. The Signature recognition systems can categorized as offline (static) and online (dynamic). This paper presents Surf Feature based recognition of offline signatures system that is trained with low-resolution scanned signature images. The signature of a person is an important biometric attribute of a human being which can be used to authenticate human identity. However the signatures of human can be handled as an image and recognized using computer vision and HMM techniques. With modern computers, there is need to develop fast algorithms for signature recognition. There are multiple techniques are defined to signature recognition with a lot of scope of research. In this paper, (static signature) off-line signature recognition & verification using surf feature with HMM is proposed, where the signature is captured and presented to the user in an image format. Signatures are verified depended on parameters extracted from the signature using various image processing techniques. The Off-line Signature Verification and Recognition is implemented using Mat lab platform. This work has been analyzed or tested and found suitable for its purpose or result. The proposed method performs better than the other recently proposed methods.

Keywords: offline signature verification, offline signature recognition, signatures, SURF features, HMM

Procedia PDF Downloads 362

Authors: Wanhyun Cho, Soonja Kang, Sangkyoon Kim, Soonyoung Park

Abstract:

In this paper, we present the human action recognition method using the variational Bayesian HMM with the Dirichlet process mixture (DPM) of the Gaussian-Wishart emission model (GWEM). First, we define the Bayesian HMM based on the Dirichlet process, which allows an infinite number of Gaussian-Wishart components to support continuous emission observations. Second, we have considered an efficient variational Bayesian inference method that can be applied to drive the posterior distribution of hidden variables and model parameters for the proposed model based on training data. And then we have derived the predictive distribution that may be used to classify new action. Third, the paper proposes a process of extracting appropriate spatial-temporal feature vectors that can be used to recognize a wide range of human behaviors from input video image. Finally, we have conducted experiments that can evaluate the performance of the proposed method. The experimental results show that the method presented is more efficient with human action recognition than existing methods.

Keywords: human action recognition, Bayesian HMM, Dirichlet process mixture model, Gaussian-Wishart emission model, Variational Bayesian inference, prior distribution and approximate posterior distribution, KTH dataset

Procedia PDF Downloads 324

Authors: Wei Chen Chang, Ming Yu Hsiao

Abstract:

The commodity takes one kind of mark, the designer how to construction and interpretation the user how to use the process and effectively convey message in design education has always been an important issue. Cultural-creative design refers to signifying cultural heritage for product design. In terms of Peirce’s Semiotic Triangle: signifying elements-object-interpretant, signifying elements are the outcomes of design, the object is cultural heritage, and the interpretant is the positioning and description of product design. How to elaborate the positioning, design, and development of a product is a narrative issue of the interpretant, and how to shape the signifying elements of a product by modifying and adapting styles is a rhetoric matter. This study investigated the rhetoric of elements signifying products to develop a rhetoric model with cultural style. Figures of speech are a rhetoric method in narrative. By adapting figures of speech to the interpretant, this study developed the rhetoric context of cultural context by narrative means. In this two-phase study, phase I defines figures of speech and phase II analyzes existing cultural-creative products in terms of figures of speech to develop a rhetoric of style model. We expect it can reference for the future development of Cultural-creative design.

Keywords: cultural-creative design, cultural-creative products, figures of speech, Peirce’s semiotic triangle, rhetoric of style model

Procedia PDF Downloads 348

Authors: Mohamed Lotfy, Ghada Soliman

Abstract:

Recognizing Arabic dot-matrix digits is a challenging problem due to the unique characteristics of dot-matrix fonts, such as irregular dot spacing and varying dot sizes. This paper presents an approach for recognizing Arabic digits printed in dot matrix format. The proposed model is based on Convolutional Neural Networks (CNN) that take the dot matrix as input and generate embeddings that are rounded to generate binary representations of the digits. The binary embeddings are then used to perform Optical Character Recognition (OCR) on the digit images. To overcome the challenge of the limited availability of dotted Arabic expiration date images, we developed a True Type Font (TTF) for generating synthetic images of Arabic dot-matrix characters. The model was trained on a synthetic dataset of 3287 images and 658 synthetic images for testing, representing realistic expiration dates from 2019 to 2027 in the format of yyyy/mm/dd. Our model achieved an accuracy of 98.94% on the expiry date recognition with Arabic dot matrix format using fewer parameters and less computational resources than traditional CNN-based models. By investigating and presenting our findings comprehensively, we aim to contribute substantially to the field of OCR and pave the way for advancements in Arabic dot-matrix character recognition. Our proposed approach is not limited to Arabic dot matrix digit recognition but can also be extended to text recognition tasks, such as text classification and sentiment analysis.

Keywords: computer vision, pattern recognition, optical character recognition, deep learning

Procedia PDF Downloads 55

Authors: Lubna Alhinti, Heidi Christensen, Stuart Cunningham

Abstract:

People with speech disorders may rely on augmentative and alternative communication (AAC) technologies to help them communicate. However, the limitations of the current AAC technologies act as barriers to the optimal use of these technologies in daily communication settings. The ability to communicate effectively relies on a number of factors that are not limited to the intelligibility of the spoken words. In fact, non-verbal cues play a critical role in the correct comprehension of messages and having to rely on verbal communication only, as is the case with current AAC technology, may contribute to problems in communication. This is especially true for people’s ability to express their feelings and emotions, which are communicated to a large part through non-verbal cues. This paper focuses on understanding more about the non-verbal communication ability of people with dysarthria, with the overarching aim of this research being to improve AAC technology by allowing people with dysarthria to better communicate emotions. Preliminary survey results are presented that gives an understanding of how people with dysarthria convey emotions, what emotions that are important for them to get across, what emotions that are difficult for them to convey, and whether there is a difference in communicating emotions when speaking to familiar versus unfamiliar people.

Keywords: alternative and augmentative communication technology, dysarthria, speech emotion recognition, VIVOCA

Procedia PDF Downloads 128

Authors: Farshideh Einsele, Hassan Foroosh

Abstract:

In this paper, we present a robust algorithm to recognize extracted text from grocery product images captured by mobile phone cameras. Recognition of such text is challenging since text in grocery product images varies in its size, orientation, style, illumination, and can suffer from perspective distortion. Pre-processing is performed to make the characters scale and rotation invariant. Since text degradations can not be appropriately defined using wellknown geometric transformations such as translation, rotation, affine transformation and shearing, we use the whole character black pixels as our feature vector. Classification is performed with minimum distance classifier using the maximum likelihood criterion, which delivers very promising Character Recognition Rate (CRR) of 89%. We achieve considerably higher Word Recognition Rate (WRR) of 99% when using lower level linguistic knowledge about product words during the recognition process.

Keywords: camera-based OCR, feature extraction, document, image processing, grocery products

Procedia PDF Downloads 379

Authors: James Rigor Camacho, Wansu Lim

Abstract:

Over the last few years, the development of new wearable and processing technologies has accelerated in order to harness physiological data such as electroencephalograms (EEGs) for EEG-based applications. EEG has been demonstrated to be a source of emotion recognition signals with the highest classification accuracy among physiological signals. However, when emotion recognition systems are used for real-time classification, the training unit is frequently left to run offline or in the cloud rather than working locally on the edge. That strategy has hampered research, and the full potential of using an edge AI device has yet to be realized. Edge AI devices are computers with high performance that can process complex algorithms. It is capable of collecting, processing, and storing data on its own. It can also analyze and apply complicated algorithms like localization, detection, and recognition on a real-time application, making it a powerful embedded device. The NVIDIA Jetson series, specifically the Jetson Nano device, was used in the implementation. The cEEGrid, which is integrated to the open-source brain computer-interface platform (OpenBCI), is used to collect EEG signals. An EEG-based real-time emotion recognition system on Edge AI is proposed in this paper. To perform graphical spectrogram categorization of EEG signals and to predict emotional states based on input data properties, machine learning-based classifiers were used. Until the emotional state was identified, the EEG signals were analyzed using the K-Nearest Neighbor (KNN) technique, which is a supervised learning system. In EEG signal processing, after each EEG signal has been received in real-time and translated from time to frequency domain, the Fast Fourier Transform (FFT) technique is utilized to observe the frequency bands in each EEG signal. To appropriately show the variance of each EEG frequency band, power density, standard deviation, and mean are calculated and employed. The next stage is to identify the features that have been chosen to predict emotion in EEG data using the K-Nearest Neighbors (KNN) technique. Arousal and valence datasets are used to train the parameters defined by the KNN technique.Because classification and recognition of specific classes, as well as emotion prediction, are conducted both online and locally on the edge, the KNN technique increased the performance of the emotion recognition system on the NVIDIA Jetson Nano. Finally, this implementation aims to bridge the research gap on cost-effective and efficient real-time emotion recognition using a resource constrained hardware device, like the NVIDIA Jetson Nano. On the cutting edge of AI, EEG-based emotion identification can be employed in applications that can rapidly expand the research and implementation industry's use.

Keywords: edge AI device, EEG, emotion recognition system, supervised learning algorithm, sensors

Procedia PDF Downloads 81

Authors: S. Jagadeesh Kumar

Abstract:

Face detection and recognition is an authoritative technology for image database management, video surveillance, and human computer interface (HCI). Face recognition is a rapidly nascent method, which has been extensively discarded in forensics such as felonious identification, tenable entree, and custodial security. This paper recommends an erudite technique using curvature analysis (CA) that has less false positives incidence, operative in different light environments and confiscates the artifacts that are introduced during image acquisition by ring correction in polar coordinate (RCP) method. This technique affronts mean and median filtering technique to remove the artifacts but it works in polar coordinate during image acquisition. Investigational fallouts for face detection and recognition confirms decent recitation even in diagonal orientation and stance variation.

Keywords: curvature analysis, ring correction in polar coordinate method, face detection, face recognition, human computer interaction

Procedia PDF Downloads 259

Authors: Ksenia Meshkova

Abstract:

With increasing amount of multimedia data, it is very important to automate and speed up the process of obtaining meta. This process means not just recognition of some object or its movement, but recognition of the entire scene versus separate frames and having timeline segmentation as a final result. Labeling datasets is time consuming, besides, attributing characteristics to particular scenes is clearly difficult due to their nature. In this article, we will consider autoencoders application to unsupervised scene recognition and clusterization based on interpretable features. Further, we will focus on particular types of auto encoders that relevant to our study. We will take a look at the specificity of deep learning related to information theory and rate-distortion theory and describe the solutions empowering poor interpretability of deep learning in media content processing. As a conclusion, we will present the results of the work of custom framework, based on autoencoders, capable of scene recognition as was deeply studied above, with highlights generation resulted out of this recognition. We will not describe in detail the mathematical description of neural networks work but will clarify the necessary concepts and pay attention to important nuances.

Keywords: neural networks, computer vision, representation learning, autoencoders

Procedia PDF Downloads 98

Authors: Gobinathan Devathasan, Kezia Devathasan

Abstract:

Objective: To improve the current modified Broca-Wernicke-Lichtheim-Kussmaul speech schema and provide insight into autism. Methods: We reviewed the pertinent literature. Current findings, involving Brodmann areas 22, 46, 9,44,45,6,4 are based on neuropathology and functional MRI studies. However, in primary autism, there is no lucid explanation and changes described, whether neuropathology or functional MRI, appear consequential. Findings: We forward an enhanced model which may explain the enigma related to autism. Vowel output is subcortical and does need cortical representation whereas consonant speech is cortical in origin. Left lateralization is needed to commence the circuitry spin as our life have evolved with L-amino acids and left spin of electrons. A fundamental species difference is we are capable of three syllable-consonants and bi-syllable expression whereas cetaceans and songbirds are confined to single or dual consonants. The 4 key sites for speech are superior auditory cortex, Broca’s two areas, and the supplementary motor cortex. Using the Argand’s diagram and Reimann’s projection, we theorize that the Euclidean three dimensional synaptic neuronal circuits of speech are quantized to coherent waves, and then decoherence takes place at area 6 (spherical representation). In this quantum state complex, 3-consonant languages are instantaneously integrated and multiple languages can be learned, verbalized and differentiated. Conclusion: We postulate that evolutionary human speech is elevated to quantum interaction unlike cetaceans and birds to achieve the three consonants/bi-syllable speech. In classical primary autism, the sudden speech switches off and on noted in several cases could now be explained not by any anatomical lesion but failure of coherence. Area 6 projects directly into prefrontal saccadic area (8); and this further explains the second primary feature in autism: lack of eye contact. The third feature which is repetitive finger gestures, located adjacent to the speech/motor areas, are actual attempts to communicate with the autistic child akin to sign language for the deaf.

Keywords: quantum neuronal paradigm, cetaceans and human speech, autism and rapid magnetic stimulation, coherence and decoherence of speech

Procedia PDF Downloads 164

Authors: Reza Shakoori

Abstract:

This paper presents a structural correction of Arabic and Persian strokes using manipulation of their chain codes in order to improve the rate and performance of Persian and Arabic handwritten word recognition systems. It collects pure and effective features to represent a character with one consolidated feature vector and reduces variations in order to decrease the number of training samples and increase the chance of successful classification. Our results also show that how the proposed approaches can simplify classification and consequently recognition by reducing variations and possible noises on the chain code by keeping orientation of characters and their backbone structures.

Keywords: Arabic, chain code normalization, OCR systems, image processing

Procedia PDF Downloads 373

Authors: Pavan K. Rallabandi, Kailash C. Patidar

Abstract:

In this paper, we present an optimization technique or a learning algorithm using the hybrid architecture by combining the most popular sequence recognition models such as Recurrent Neural Networks (RNNs) and Hidden Markov models (HMMs). In order to improve the sequence or pattern recognition/ classification performance by applying a hybrid/neural symbolic approach, a gradient descent learning algorithm is developed using the Real Time Recurrent Learning of Recurrent Neural Network for processing the knowledge represented in trained Hidden Markov Models. The developed hybrid algorithm is implemented on automata theory as a sample test beds and the performance of the designed algorithm is demonstrated and evaluated on learning the deterministic finite state automata.

Keywords: hybrid systems, hidden markov models, recurrent neural networks, deterministic finite state automata

Procedia PDF Downloads 360

Authors: Sheena Christabel Pravin, M. Palanivelan

Abstract:

Speech disfluencies are common in spontaneous speech. The primary purpose of this study was to distinguish linguistic disfluencies from stuttering disfluencies in bilingual Tamil–English (TE) speaking children. The secondary purpose was to determine whether their disfluencies are mediated by native language dominance and/or on an early onset of developmental stuttering at childhood. A detailed study was carried out to identify the prosodic and acoustic features that uniquely represent the disfluent regions of speech. This paper focuses on statistical modeling of repetitions, prolongations, pauses and interjections in the speech corpus encompassing bilingual spontaneous utterances from school going children – English and Tamil. Two classifiers including Hidden Markov Models (HMM) and the Multilayer Perceptron (MLP), which is a class of feed-forward artificial neural network, were compared in the classification of disfluencies. The results of the classifiers document the patterns of disfluency in spontaneous speech samples of school-aged children to distinguish between Children Who Stutter (CWS) and Children with Language Impairment CLI). The ability of the models in classifying the disfluencies was measured in terms of F-measure, Recall, and Precision.

Keywords: bi-lingual, children who stutter, children with language impairment, hidden markov models, multi-layer perceptron, linguistic disfluencies, stuttering disfluencies

Procedia PDF Downloads 194

Authors: Rodolfo Lorbieski, Silvia Modesto Nassar

Abstract:

Many supervised machine learning tasks require decision making across numerous different classes. Multi-class classification has several applications, such as face recognition, text recognition and medical diagnostics. The objective of this article is to analyze an adapted method of Stacking in multi-class problems, which combines ensembles within the ensemble itself. For this purpose, a training similar to Stacking was used, but with three levels, where the final decision-maker (level 2) performs its training by combining outputs from the tree-based pair of meta-classifiers (level 1) from Bayesian families. These are in turn trained by pairs of base classifiers (level 0) of the same family. This strategy seeks to promote diversity among the ensembles forming the meta-classifier level 2. Three performance measures were used: (1) accuracy, (2) area under the ROC curve, and (3) time for three factors: (a) datasets, (b) experiments and (c) levels. To compare the factors, ANOVA three-way test was executed for each performance measure, considering 5 datasets by 25 experiments by 3 levels. A triple interaction between factors was observed only in time. The accuracy and area under the ROC curve presented similar results, showing a double interaction between level and experiment, as well as for the dataset factor. It was concluded that level 2 had an average performance above the other levels and that the proposed method is especially efficient for multi-class problems when compared to binary problems.

Keywords: stacking, multi-layers, ensemble, multi-class

Procedia PDF Downloads 245

Authors: Jonathan Sohn

Abstract:

Deep neural networks (DNNs) have shown state-of-the-art performance for many applications, including computer vision, natural language processing, and speech recognition. Recently, adversarial attacks have been studied in the context of deep neural networks, which aim to alter the results of deep neural networks by modifying the inputs slightly. For example, an adversarial attack on a DNN used for object detection can cause the DNN to miss certain objects. As a result, the reliability of DNNs is undermined by their lack of robustness against adversarial attacks, raising concerns about their use in safety-critical applications such as autonomous driving. In this paper, we focus on studying the adversarial attacks and defenses on DNNs for image classification. There are two types of adversarial attacks studied which are fast gradient sign method (FGSM) attack and projected gradient descent (PGD) attack. A DNN forms decision boundaries that separate the input images into different categories. The adversarial attack slightly alters the image to move over the decision boundary, causing the DNN to misclassify the image. FGSM attack obtains the gradient with respect to the image and updates the image once based on the gradients to cross the decision boundary. PGD attack, instead of taking one big step, repeatedly modifies the input image with multiple small steps. There is also another type of attack called the target attack. This adversarial attack is designed to make the machine classify an image to a class chosen by the attacker. We can defend against adversarial attacks by incorporating adversarial examples in training. Specifically, instead of training the neural network with clean examples, we can explicitly let the neural network learn from the adversarial examples. In our experiments, the digit recognition accuracy on the MNIST dataset drops from 97.81% to 39.50% and 34.01% when the DNN is attacked by FGSM and PGD attacks, respectively. If we utilize FGSM training as a defense method, the classification accuracy greatly improves from 39.50% to 92.31% for FGSM attacks and from 34.01% to 75.63% for PGD attacks. To further improve the classification accuracy under adversarial attacks, we can also use a stronger PGD training method. PGD training improves the accuracy by 2.7% under FGSM attacks and 18.4% under PGD attacks over FGSM training. It is worth mentioning that both FGSM and PGD training do not affect the accuracy of clean images. In summary, we find that PGD attacks can greatly degrade the performance of DNNs, and PGD training is a very effective way to defend against such attacks. PGD attacks and defence are overall significantly more effective than FGSM methods.

Keywords: deep neural network, adversarial attack, adversarial defense, adversarial machine learning

Procedia PDF Downloads 162

Authors: Lanchi Xie, Zhihui Li, Zhigang Li, Guiqiang Wang, Lei Xu, Yuwen Yan

Abstract:

Image-based facial features can be classified into category recognition features and individual recognition features. Current automated face recognition systems extract a specific feature vector of different dimensions from a facial image according to their pre-trained neural network. However, to improve the efficiency of parameter calculation, an algorithm generally reduces the image details by pooling. The operation will overlook the details concerned much by forensic experts. In our experiment, we adopted a variety of face recognition algorithms based on deep learning, compared a large number of naturally collected face images with the known data of the same person's frontal ID photos. Downscaling and manual handling were performed on the testing images. The results supported that the facial recognition algorithms based on deep learning detected structural and morphological information and rarely focused on specific markers such as stains and moles. Overall performance, distribution of genuine scores and impostor scores, and likelihood ratios were tested to evaluate the accuracy of biometric systems and forensic experts. Experiments showed that the biometric systems were skilled in distinguishing category features, and forensic experts were better at discovering the individual features of human faces. In the proposed approach, a fusion was performed at the score level. At the specified false accept rate, the framework achieved a lower false reject rate. This paper contributes to improving the interpretability of the objective method of facial comparison and provides a novel method for human-machine collaboration in this field.

Keywords: likelihood ratio, automated facial recognition, facial comparison, biometrics

Procedia PDF Downloads 103

Authors: Kilari Nikhil, Ankur Tibrewal, Srinivas Kruthiventi S. S.

Abstract:

Conventional speaker identification systems often fall short of capturing the diverse variations present in speech data due to fixed-scale architectures. In this research, we propose a CNN-based architecture, USENet, designed to overcome these limitations. Leveraging two key techniques, our approach achieves superior performance on the VoxCeleb 1 Dataset without any pre-training. Firstly, we adopt a U-net-inspired design to extract features at multiple scales, empowering our model to capture speech characteristics effectively. Secondly, we introduce the squeeze and excitation block to enhance spatial feature learning. The proposed architecture showcases significant advancements in speaker identification, outperforming existing methods, and holds promise for future research in this domain.

Keywords: multi-scale feature extraction, squeeze and excitation, VoxCeleb1 speaker identification, mel-spectrograms, USENet

Procedia PDF Downloads 46

Authors: Qin Long, Li Xiaoge

Abstract:

The Knowledge Graphs have now become a new form of knowledge representation. However, there is no consensus in regard to a plausible and definition of entities and relationships in the domain-specific knowledge graph. Further, in conjunction with several limitations and deficiencies, various domain-specific entities and relationships recognition approaches are far from perfect. Specifically, named entity recognition in Chinese domain is a critical task for the natural language process applications. However, a bottleneck problem with Chinese named entity recognition in new domains is the lack of annotated data. To address this challenge, a domain distant supervised named entity recognition framework is proposed. The framework is divided into two stages: first, the distant supervised corpus is generated based on the entity linking model of graph attention neural network; secondly, the generated corpus is trained as the input of the distant supervised named entity recognition model to train to obtain named entities. The link model is verified in the ccks2019 entity link corpus, and the F1 value is 2% higher than that of the benchmark method. The re-pre-trained BERT language model is added to the benchmark method, and the results show that it is more suitable for distant supervised named entity recognition tasks. Finally, it is applied in the computer field, and the results show that this framework can obtain domain named entities.

Keywords: distant named entity recognition, entity linking, knowledge graph, graph attention neural network

Procedia PDF Downloads 71

Authors: Shrabani Ghosh

Abstract:

Automatic offensive language detection in social media has become a stirring task in today's NLP. Manual Offensive language detection is tedious and laborious work where automatic methods based on machine learning are only alternatives. Previous works have done sentiment analysis over social media in different ways such as supervised, semi-supervised, and unsupervised manner. Domain adaptation in a semi-supervised way has also been explored in NLP, where the source domain and the target domain are different. In domain adaptation, the source domain usually has a large amount of labeled data, while only a limited amount of labeled data is available in the target domain. Pretrained transformers like BERT, RoBERTa models are fine-tuned to perform text classification in an unsupervised manner to perform further pre-train masked language modeling (MLM) tasks. In previous work, hate speech detection has been explored in Gab.ai, which is a free speech platform described as a platform of extremist in varying degrees in online social media. In domain adaptation process, Twitter data is used as the source domain, and Gab data is used as the target domain. The performance of domain adaptation also depends on the cross-domain similarity. Different distance measure methods such as L2 distance, cosine distance, Maximum Mean Discrepancy (MMD), Fisher Linear Discriminant (FLD), and CORAL have been used to estimate domain similarity. Certainly, in-domain distances are small, and between-domain distances are expected to be large. The previous work finding shows that pretrain masked language model (MLM) fine-tuned with a mixture of posts of source and target domain gives higher accuracy. However, in-domain performance of the hate classifier on Twitter data accuracy is 71.78%, and out-of-domain performance of the hate classifier on Gab data goes down to 56.53%. Recently self-supervised learning got a lot of attention as it is more applicable when labeled data are scarce. Few works have already been explored to apply self-supervised learning on NLP tasks such as sentiment classification. Self-supervised language representation model ALBERTA focuses on modeling inter-sentence coherence and helps downstream tasks with multi-sentence inputs. Self-supervised attention learning approach shows better performance as it exploits extracted context word in the training process. In this work, a self-supervised attention mechanism has been proposed to detect hate speech on Gab.ai. This framework initially classifies the Gab dataset in an attention-based self-supervised manner. On the next step, a semi-supervised classifier trained on the combination of labeled data from the first step and unlabeled data. The performance of the proposed framework will be compared with the results described earlier and also with optimized outcomes obtained from different optimization techniques.

Keywords: attention learning, language model, offensive language detection, self-supervised learning

Procedia PDF Downloads 83

Authors: Xharavina V., Gallopeni F., Ahmeti K.

Abstract:

Stuttered speech is filled with intermittent sound prolongations and/or rapid part word repetitions. Oftentimes, these aberrant acoustic behaviors are associated with intermittent physical tension and struggle behaviors such as head jerks, arm jerks, finger tapping, excessive eye-blinks, etc. Additionally, the jarring nature of acoustic and physical manifestations that often accompanies moderate-severe stuttering may induce negative emotional responses in listeners, which alters communication between the person who stutters and their listeners. However, researches for the influence of negative emotions in the communication and for physical reaction are limited. Therefore, to compare psycho-physiological responses of fluent adults, while listening the speech of adults who speak fluency and adults who stutter, are necessary. This study comprises the experimental method, with total of 104 participants (average age-20 years old, SD=2.1), divided into 3 groups. All participants self-reported no impairments in speech, language, or hearing. Exploring the responses of the participants, there were used two records speeches; a voice who speaks fluently and the voice who stutters. Heartbeats and the pulse were measured by the digital blood pressure monitor called 'Tensoval', as a physiological response to the fluent and stuttering sample. Meanwhile, the emotional responses of participants were measured by the self-reporting questionnaire (Steenbarger, 2001). Results showed an increase in heartbeats during the stuttering speech compared with the fluent sample (p < 0.5). The listeners also self-reported themselves as more alive, unhappy, nervous, repulsive, sad, tense, distracted and upset when listening the stuttering words versus the words of the fluent adult (where it was reported to experience positive emotions). These data support the notions that speech with stuttering can bring a psycho-physical reaction to the listeners. Speech pathologists should be aware that listeners show intolerable physiological reactions to stuttering that remain visible over time.

Keywords: emotional, physiological, stuttering, fluent speech

Procedia PDF Downloads 122

Authors: Stefania Arguelles Reyes, Octavio José Salcedo Parra, Alberto Acosta López

Abstract:

This paper presents a web application for the improvement of images through recognition. The web application is based on the analysis of picture-based recognition methods that allow an improvement on the physical appearance of people posting in social networks. The basis relies on the study of tools that can correct or improve some features of the face, with the help of a wide collection of user images taken as reference to build a facial profile. Automatic facial profiling can be achieved with a deeper study of the Object Detection Library. It was possible to improve the initial images with the help of MATLAB and its filtering functions. The user can have a direct interaction with the program and manually adjust his preferences.

Keywords: Matlab, make up, recognition methods, web application

Procedia PDF Downloads 116

Authors: Zoran Spasovski

Abstract:

The paper shortly discusses the significance and the benefits of the historical approach in the research of languages by presenting examples of it in the fields of phonetics and phonology, lexicology, morphology, syntax, and even in the onomastics (toponomy and anthroponomy). The examples from the field of phonetics/phonology include insights into animal speech and its evolution into human speech, the evolution of the sounds of human speech from vocals to glides and consonants and from velar consonants to palatal, etc., on well-known examples of former researchers. Those from the field of lexicology show shortly the formation of the lexemes and their evolution; the morphology and syntax are explained by examples of the development of grammar and syntax forms, and the importance of the historical approach in the research of place-names and personal names is briefly outlined through examples of place-names and personal names and surnames, and the conclusions that come from it, in different languages.

Keywords: animal speech, glotogenesis, grammar forms, lexicology, place-names, personal names, surnames, syntax categories

Procedia PDF Downloads 49

Authors: Wale Agbaje

Abstract:

The aim of this research study is to assess the impact of financial statement fraud on profitability of some selected Nigerian manufacturing firms covering (2002-2016). The specific objectives focused on to ascertain the effect of incorrect asset valuation on return on assets (ROA) and to ascertain the relationship between improper expense recognition and return on assets (ROA). To achieve these objectives, descriptive research design was used for the study while secondary data were collected from the financial reports of the selected firms and website of security and exchange commission. The analysis of covariance (ANCOVA) was used and STATA II econometric method was used in the analysis of the data. Altman model and operating expenses ratio was adopted in the analysis of the financial reports to create a dummy variable for the selected firms from 2002-2016 and validation of the parameters were ascertained using various statistical techniques such as t-test, co-efficient of determination (R2), F-statistics and Wald chi-square. Two hypotheses were formulated and tested using the t-statistics at 5% level of significance. The findings of the analysis revealed that there is a significant relationship between financial statement fraud and profitability in Nigerian manufacturing industry. It was revealed that incorrect assets valuation has a significant positive relationship and so also is the improper expense recognition on return on assets (ROA) which serves as a proxy for profitability. The implication of this is that distortion of asset valuation and expense recognition leads to decreasing profit in the long run in the manufacturing industry. The study therefore recommended that pragmatic policy options need to be taken in the manufacturing industry to effectively manage incorrect asset valuation and improper expense recognition in order to enhance manufacturing industry performance in the country and also stemming of financial statement fraud should be adequately inculcated into the internal control system of manufacturing firms for the effective running of the manufacturing industry in Nigeria.

Keywords: Althman's Model, improper expense recognition, incorrect asset valuation, return on assets

Procedia PDF Downloads 133

Authors: Frederik Calsius, Mirela Popa, Alexia Briassouli

Abstract:

In the field of machine learning, it is common practice to use benchmark datasets to prove the working of a method. The domain of action recognition in videos often uses datasets like Kinet-ics, Something-Something, UCF-101 and HMDB-51 to report results. Considering the properties of the datasets, there are no datasets that focus solely on very short clips (2 to 3 seconds), and on highly-similar fine-grained actions within one specific domain. This paper researches how current state-of-the-art action recognition methods perform on a dataset that consists of highly similar, fine-grained actions. To do so, a dataset of skateboarding tricks was created. The performed analysis highlights both benefits and limitations of state-of-the-art methods, while proposing future research directions in the activity recognition domain. The conducted research shows that the best results are obtained by fusing RGB data with OpenPose data for the Temporal Shift Module.

Keywords: activity recognition, fused deep representations, fine-grained dataset, temporal modeling

Procedia PDF Downloads 204