Recognition Engine Research Articles

Convolutional neural network with joint stepwise character/word modeling based system for scene text recognition

Text recognition in the wild is a challenging task in the field of computer vision and machine learning. Existing optical character recognition engines cannot perform well in the natural scene. In this context, deep learning models have emerged as a powerful state-of-the-art technique in the classification and recognition process. This study proposes a new Convolutional Neural Network based system for scene text reading. We investigate how to combine the character recognition module followed by the word recognition module to achieve the overall system goal. The first module analyzes characters within multi-scale images by relaying on the power of the convolutional network and the fully connected network for character recognition. The second module relies on the Viterbi search to find the closest word to a given characters sequence. For the sake of more precision, a bigram based linguistic module is applied. The proposed system achieves the state-of-the-art performance on three standard scene text recognition benchmarks: chars74k, ICDAR 2003 and ICDAR 2013. In particular, this performance is proven on both of character and word recognition accuracy as well as speed aspects via a comparative study between different deep learning architectures.

Performance of Forced-Alignment Algorithms on Children's Speech.

Purpose Acoustic measurement of speech sounds requires first segmenting the speech signal into relevant units (words, phones, etc.). Manual segmentation is cumbersome and time consuming. Forced-alignment algorithms automate this process by aligning a transcript and a speech sample. We compared the phoneme-level alignment performance of five available forced-alignment algorithms on a corpus of child speech. Our goal was to document aligner performance for child speech researchers. Method The child speech sample included 42 children between 3 and 6 years of age. The corpus was force-aligned using the Montreal Forced Aligner with and without speaker adaptive training, triphone alignment from the Kaldi speech recognition engine, the Prosodylab-Aligner, and the Penn Phonetics Lab Forced Aligner. The sample was also manually aligned to create gold-standard alignments. We evaluated alignment algorithms in terms of accuracy (whether the interval covers the midpoint of the manual alignment) and difference in phone-onset times between the automatic and manual intervals. Results The Montreal Forced Aligner with speaker adaptive training showed the highest accuracy and smallest timing differences. Vowels were consistently the most accurately aligned class of sounds across all the aligners, and alignment accuracy increased with age for fricative sounds across the aligners too. Conclusion The best-performing aligner fell just short of human-level reliability for forced alignment. Researchers can use forced alignment with child speech for certain classes of sounds (vowels, fricatives for older children), especially as part of a semi-automated workflow where alignments are later inspected for gross errors. Supplemental Material https://doi.org/10.23641/asha.14167058.

Automatic Speech Recognition for Indoor HRI Scenarios

This article presents a stand-alone automatic speech recognition system that accounts for listener movement, time-varying reverberation effects, environmental noise, and user position information for beamforming approaches in an HRI setting. We raise the importance of replacing the classical black-box integration of automatic speech recognition technology in HRI applications with the incorporation of the acoustic environment representation and modeling, and of the target source direction. Test data were recorded on a real robot under various moving conditions. For addressing the time-varying acoustic channel problem and incorporating environmental effect during training, clean speech samples were passed through estimated static channel responses and noise was added. Beamforming is investigated regarding oracle source tracking using, for instance, image processing. The proposed strategy is interesting for the robotics community, because it allows the development of voice-based HRI with limited training data and without relying on third-party technologies or Internet access eliminating the need to upload data to the cloud. In our mobile HRI scenario, the resulting speech recognition engine provided an average word error rate that is at least 19% and 34% lower than publicly available speech recognition APIs with the playback (i.e., loudspeaker) and human testing modalities, respectively.

Antibiotics as Inhibitor of Glutathione S-transferase: Biological Evaluation and Molecular Structure Studies.

The glutathione S-transferases (GSTs) are family of enzymes that are notable for their role in phase II detoxification reactions. Antibiotics have been reported to have several adverse effects on the activity of the enzymes in mammals. The aim of this study was the structural and biochemical characterization of rat erythrocyte GST and understanding the effects of gentamicin, clindamycin, cefazolin, ampicillin and scopolamine butylbromide on the activity of human erythrocyte GST using rat as a model. The enzyme was purified by GSH-agarose affinity chromatography. In vitro GST enzyme activity was measured at 25°C using CDNB as a model substrate. IC50 of drugs was measured by activity % vs compound concentration graphs. Lineweaver Burk graphs were drawn to determine the inhibition type and Ki constants for the drugs. The structure of the enzyme was predicted via Protein Homology/analogy Recognition Engine. In this study, GST was purified from rat erythrocyte with a specific activity of 6.3 EU/mg protein, 44 % yield and 115 fold. Gentamicin and clindamycin inhibited the enzymatic activity with IC50 of 1.69 and 6.9 mM and Ki of 1.70 and 2.36 mM, respectively. Ampicillin and scopolamine butylbromide were activators of the enzyme, while the activity of the enzyme was insensitive to cefazolin. The enzyme was further characterized by homology modeling and sequence alignment revealing similarities with human GST. Collectively, it could be concluded that gentamicin and clindamycin are the inhibitors of erythrocyte GST.

RETRACTED: Design of english translation platform based on embedded system software simulation

English translation platform system design, the system is configured to perform cross language communication between spoken on public handheld devices. To-end system using the n-gram large vocabulary speech recognition engine to identify and said English. This phenomenon indicates that there are some problems in the whole system. These problems may be due to the nonstandard English problems or some instructions are not expressed in standard English.All in all, these problems are caused by the inaccurate translation.There are also some other translation engines that can translate all kinds of foreign languages, and even transfer the translated text directly to the speech synthesizer for voice display.This system can not only describe the functions and components of the system, but also realize a real-time voice simulation on some low-power platforms.The simulation of this language has been applied in many fields by many scientists. Simulation is used to simulate the behavior of some systems, so as to estimate, predict and modify the behavior.The vehicle virtual collision test proposed in this research is a kind of simulation technology.Using virtual car collision to improve the safety of the car, there is no need to carry out experiments in real life, thus reducing the cost of the experiment, but also reducing the difficulty of the experiment.

A Hybrid Swarm and Gravitation-based feature selection algorithm for handwritten Indic script classification problem

In any multi-script environment, handwritten script classification is an unavoidable pre-requisite before the document images are fed to their respective Optical Character Recognition (OCR) engines. Over the years, this complex pattern classification problem has been solved by researchers proposing various feature vectors mostly having large dimensions, thereby increasing the computation complexity of the whole classification model. Feature Selection (FS) can serve as an intermediate step to reduce the size of the feature vectors by restricting them only to the essential and relevant features. In the present work, we have addressed this issue by introducing a new FS algorithm, called Hybrid Swarm and Gravitation-based FS (HSGFS). This algorithm has been applied over three feature vectors introduced in the literature recently—Distance-Hough Transform (DHT), Histogram of Oriented Gradients (HOG), and Modified log-Gabor (MLG) filter Transform. Three state-of-the-art classifiers, namely, Multi-Layer Perceptron (MLP), K-Nearest Neighbour (KNN), and Support Vector Machine (SVM), are used to evaluate the optimal subset of features generated by the proposed FS model. Handwritten datasets at block, text line, and word level, consisting of officially recognized 12 Indic scripts, are prepared for experimentation. An average improvement in the range of 2–5% is achieved in the classification accuracy by utilizing only about 75–80% of the original feature vectors on all three datasets. The proposed method also shows better performance when compared to some popularly used FS models. The codes used for implementing HSGFS can be found in the following Github link: https://github.com/Ritam-Guha/HSGFS.

Content-based Image Retrieval using Tesseract OCR Engine and Levenshtein Algorithm

Image Retrieval Systems (IRSs) are applications that allow one to retrieve images saved at any location on a network. Most IRSs make use of reverse lookup to find images stored on the network based on image properties such as size, filename, title, color, texture, shape, and description. This paper provides a technique for obtaining full image document given that the user has some portions of the document under search. To demonstrate the reliability of the proposed technique, we designed a system to implement the algorithm. A combination of Optical Character Recognition (OCR) engine and an improved text-matching algorithm was used in the system implementation. The Tesseract OCR engine and Levenshtein Algorithm was integrated to perform the image search. The extracted text is compared to the text stored in the database. For example, a query result is returned when a significant ratio of 0.15 and above is obtained. The results showed a 100% successful retrieval of the appropriate file base on the match even when partial query images were submitted.

Spoken Turkish in television news and debates: Some acoustic and morphological aspects relevant to respeaking

This paper presents a qualitative analysis of spoken Turkish based on approximately 77 minutes of transcribed content from television news and debates with special focus on acoustic and morphological aspects relevant to respeaking, where an edited form of the original verbal content is dictated to a speaker-dependent automatic speech recognition (ASR) engine, whose output is further edited for broadcast-quality subtitles. The data suggest that respeaking can solve only some of the potential problems unscripted speech presents for ASR. On the acoustic level, a respeaker can overcome segmental and suprasegmental variation as well as degraded acoustic conditions, and can partially resolve overlapping speech. On the morphological level, disfluency and deviant morphology can be handled. When paraphrasing the text and dictating punctuation marks under time pressure, a respeaker can hardly control her own pronunciation; deletion, reduced morphemes, etc., might lead to misrecognition. As standard orthography, capitalization, and punctuation are required in subtitles, named entities, figures, morphophonemics and use of apostrophe will require satisfactory solutions during ASR and/or post-editing.

SMILE: A verbal and graphical user interface tool for speech-control of soccer robots in Ghana

SMILE (Smartphone Intuitive Likeness and Engagement) application, a portable Android application that allows a human to control a robot using speech input. SMILE is a novel open source and platform independent tool that will contribute to the robot soccer research by allowing robot handlers to verbally command robots. The application resides on a smartphone embedded in the face of a humanoid robot, using a speech recognition engine to analyze user speech input while using facial expressions and speech generation to express comprehension feedback to the user. With the introduction of intuitive human robot interaction into the arena of robot soccer, we discuss a couple specific scenarios in which SMILE could improve both the pace of the game and autonomous appearance of the robots. The ability of humans to communicate verbally is essential for any cooperative task, especially fast-paced sports. In the game of soccer, players must speak with coaches, referees, and other players on either team. Therefore, if humanoids are expected to compete on the same playing field as elite soccer players in the near future, then we must expect them to be treated like humans, which include the ability to listen and converse. SMILE (Smartphone Intuitive Likeness and Engagement) is the first platform independent smartphone based tool to equip robots with these capabilities. Currently, humanoid soccer research is heavily focused on walking dynamics, computer vision, and intelligent systems; however human-robot interaction (HRI) is overlooked. We delved into this area of robot soccer by implementing SMILE, an Android application that sends data packets to the robot's onboard computer upon verbal interaction with a user.

Application of Meta Learning in Face Recognition

Now is the information age, artificial intelligence rises under the background of this era, but in the field of artificial intelligence, deep learning stands out in machine learning with its excellent learning ability and is rapidly applied to face recognition engineering. However, the success of mainstream deep learning often depends on a lot of training data and training time. This paper proposes to use meta-learning technology to realize face recognition. metalearning is to use the prior knowledge and experience to guide the learning of new tasks and has the ability to learn so that it can avoid the phenomenon of overfitting in the case of only a small number of samples. We regard the data set as a task, that is, meta-train task and meta-test task, using the MAML algorithm [1] fine-tuning gradient model based on prior knowledge to obtain optimization parameters to realize face recognition. Experiments show that deep learning neural networks using meta-learning technology can achieve higher accuracy and rate than ordinary face recognition neural networks. In the experiment, the recognition rate of the face can reach more than 92.6%. It has more intelligence and has made a contribution to the development of face recognition technology.

A speech recognition approach for an industrial training station

This paper presents a speech recognition service used in the context of commanding and guiding the activities around an industrial training station. The entire concept is built on a decentralized microservice architecture and one of the many hardware and software components is the speech recognition engine. This engine grants users the possibility to interact seamlessly with other components in order to ensure a gradual and productive learning process. By working with different API’s for both English and Romanian languages, the presented approach manages to obtain good speech recognition for defining task phrases aiding the training procedure and to reduce the recognition required time by almost 50%.

Mutational sensitivity of D614G in spike protein of SARS-CoV-2 in Jordan

BackgroundSpike protein is the surface glycoprotein of the severe acute respiratory syndrome-coronavirus-2 (SARS-CoV-2) necessary for the entry of the virus via the transmembrane receptors of the human respiratory cells causing COVID-19 disease. AimHere, we aimed to predict the three-dimensional monomer structure of spike protein of SARS-CoV-2 from 20 Jordanian nasopharyngeal samples and to determine the percentage of single amino acid variants (SAV) in the spike protein of SARS-CoV-2. MethodsThe output of the Protein Homology/analogY Recognition Engine V 2.0 (Phyre2) found four single amino acid variants in the spike gene. ResultsThe first variant represented by 5% of samples that showed tyrosine deletion at Y144 located in the N terminal domain. The second and the dominant variant, represented by 62%, showed aspartate a coil amino acid substitution to glycine an extracellular amino acid at D614G located in the spike recognition binding site. The third variant, represented by 5%, showed aspartate substitution to tyrosine at D1139Y, and the fourth variant, represented by 5% glycine substitution to serine at G1167S. ConclusionOur results have shown low mutational sensitivity in all variants except to D614G the one with the most likely neutral mutational sensitivity that all variants might not explicitly affect the function of spike glycoprotein. However, D614G might change the viral conformational plasticity and hence a potential viral fitness gain but one must be cautious about drawing any concrete conclusions about the severity of symptoms and viral transmission from genomic data only. General significanceStudying mutations such as D614G in deep is essential to control the pandemic in terms of immune systems, antibodies, or even vaccines.

Accurate Physical Activity Recognition using Multidimensional Features and Markov Model for Smart Health Fitness

Recent developments in sensor technologies enable physical activity recognition (PAR) as an essential tool for smart health monitoring and for fitness exercises. For efficient PAR, model representation and training are significant factors contributing to the ultimate success of recognition systems because model representation and accurate detection of body parts and physical activities cannot be distinguished if the system is not well trained. This paper provides a unified framework that explores multidimensional features with the help of a fusion of body part models and quadratic discriminant analysis which uses these features for markerless human pose estimation. Multilevel features are extracted as displacement parameters to work as spatiotemporal properties. These properties represent the respective positions of the body parts with respect to time. Finally, these features are processed by a maximum entropy Markov model as a recognition engine based on transition and emission probability values. Experimental results demonstrate that the proposed model produces more accurate results compared to the state-of-the-art methods for both body part detection and for physical activity recognition. The accuracy of the proposed method for body part detection is 90.91% on a University of Central Florida’s (UCF) sports action dataset and, for activity recognition on a UCF YouTube action dataset and an IM-DailyRGBEvents dataset, accuracy is 89.09% and 88.26% respectively.

Stochastic Recognition of Physical Activity and Healthcare Using Tri-Axial Inertial Wearable Sensors

The classification of human activity is becoming one of the most important areas of human health monitoring and physical fitness. With the use of physical activity recognition applications, people suffering from various diseases can be efficiently monitored and medical treatment can be administered in a timely fashion. These applications could improve remote services for health care monitoring and delivery. However, the fixed health monitoring devices provided in hospitals limits the subjects’ movement. In particular, our work reports on wearable sensors that provide remote monitoring that periodically checks human health through different postures and activities to give people timely and effective treatment. In this paper, we propose a novel human activity recognition (HAR) system with multiple combined features to monitor human physical movements from continuous sequences via tri-axial inertial sensors. The proposed HAR system filters 1D signals using a notch filter that examines the lower/upper cutoff frequencies to calculate the optimal wearable sensor data. Then, it calculates multiple combined features, i.e., statistical features, Mel Frequency Cepstral Coefficients, and Gaussian Mixture Model features. For the classification and recognition engine, a Decision Tree classifier optimized by the Binary Grey Wolf Optimization algorithm is proposed. The proposed system is applied and tested on three challenging benchmark datasets to assess the feasibility of the model. The experimental results show that our proposed system attained an exceptional level of performance compared to conventional solutions. We achieved accuracy rates of 88.25%, 93.95%, and 96.83% over MOTIONSENSE, MHEALTH, and the proposed self-annotated IM-AccGyro human-machine dataset, respectively.

Assistive Mobile Application for Visually Impaired People

According to the World Health Organization estimation, globally the number of people with some visual impairment is estimated to be 285 million, of whom 39 million are blind. The inability to use features such as sending and reading of email, schedule management, pathfinding or outdoor navigation, and reading SMS is a disadvantage for blind people in many professional and educational situations. Speech or text analysis can help improve support for visually-impaired people. Users can speak a command to perform a task. The spoken command will be interpreted by the Speech Recognition Engine (SRE) and can be converted into text or perform suitable actions. In this paper, an application that allows schedule management, emailing, and SMS reading completely based on voice command is proposed, implemented, and validated. The System hopes to provide blind people to simply speak the desired functionality and be guided thereby the system’s audio instructions. The proposed and designed app is implemented to support three languages which are English, Hindi, and Kannada.

Attention-Sharing Initiative of Multimodal Processing in Simultaneous Interpreting

This study sets out to describe simultaneous interpreters' attention-sharing initiatives when exposed under input from both videotaped speech recording and real-time transcriptions. Separation of mental energy in acquiring visual input accords with the human brain's statistic optimization principle where the same property of an object is presented through diverse fashions. In examining professional interpreters' initiatives, the authors invited five professional English-Chinese conference interpreters to simultaneously interpret a videotaped speech with real-time captions generated by speech recognition engine while meanwhile monitoring their eye movements. The results indicate the professional interpreters' preferences in referring to visually presented captions along with the speaker's facial expressions, where low-frequency words, proper names, and numbers gained greater attention than words with higher frequency. This phenomenon might be explained by the working memory theory in which the central executive enables redundancy gains retrieved from dual-channel information.

Hierarchical classifier design for speech emotion recognition in the mixed-cultural environment

ABSTRACT Recognition of emotion in speech is a difficult task due to many speaker factors like gender, age, and the cultural background (nationality, ethnicity, and region) as well as the acoustical environment. Among these factors, the cultural background of the speaker has a strong influence on the expression of emotion. The reason for the unsatisfactory performance of an emotion recognition engine built using mixed-cultural samples can be traced back to this. To address this issue, a two-level hierarchical engine has been designed to identify emotion from the speech of different cultural backgrounds. The first level of the hierarchical engine is a culture identification system, which identifies the corpus of an input utterance. As most of the speakers involved in the construction of a specific corpus are from the same locality and cultural background, we assume that a corpus represents the cultural background of the speakers of the corpus constructed. Based on the response of the first level classifier, the input utterance is forwarded to an appropriate corpus-specific emotion recognition engine, in the second level. Each corpus-specific emotion recognition system is a discriminative, multiclass SVM classifier, trained with the emotional utterances of that particular corpus. The system has been tested with five different corpora, collected from diverse cultural backgrounds, namely EMO-DB, SAVEE, IITKGP-SEC, Spanish corpus S0329, and CMU’s Woogles corpus. The system achieved an accuracy of 82.01% which is an improvement of 13.38% over monolithic approaches.

Balinese character recognition on mobile application based on tesseract open source OCR engine

Balinese script is a part of Balinese culture is rarely used today. The Provincial Government of Bali with the Governor Regulation number 80 of 2018 is trying to preserve the Balinese language and script. This study aimed at preserving the Balinese script through a mobile technology approach which is the recent trend with worldwide coverage for supporting ubiquitous learning. This research integrated the Android application to recognize Balinese characters in the form of images into text with Tesseract open source Optical Character Recognition (OCR) engine. The input of this application is a Balinese script image captured by a mobile camera or from a Balinese script image. The application recognized input image into text that can be further processed based on training data available in the application. The new Balinese script training data was created based on eighteen Balinese script’s basic syllables and numbers only. This application can be operated offline with mobile hardware that supports camera functions. The result for testing for 50-word, recognition was 62% obtained in good quality image-based Bali-Simbar font. This application can be further developed to recognize other character repertoire i.e., vowels (Akśara Suara), semi vowels (Arda Suara), additional syllables (Akśara Şwalalita), and sound killers (Pangangge Tengenan).

In silico analysis of the structural diversity and interactions between invertases and invertase inhibitors from potato (Solanum tuberosum L.).

We performed sequence diversity, phylogenetic profiling, 3D structure modelling and in silico interactions between invertases (cell wall/apoplastic and vacuolar) and invertase inhibitors (cell wall/apoplastic and vacuolar) from potato. Cloning and sequencing of invertase inhibitors was performed from different potato cultivars. The comparison of the protein sequences of the different isoforms of invertases and invertase inhibitors exhibited insertions and deletions as well as the variation in terms of amino acid residues. Furthermore, the phylogenetic tree analysis displayed two groups of invertase inhibitors corresponding to the cell wall/apoplast and vacuole. Using Phyre2 protein homology recognition engine, it revealed that the structure of invertase inhibitors was predominantly α-helical and that of invertase was α helices and β strands. Results of the Ramachandran plots for each structure showed that the percentage of amino acid residues in favoured region and in allowed region. Also, the Z score and QMEAN score indicated overall good, acceptable and reliable models. In silico interactions between different isoforms of invertase and invertase inhibitors suggested that cell wall/ apoplastic invertase inhibitor exhibited stronger interaction with vacuolar invertase compared to the vacuolar invertase inhibitor. In silico interactions provides valuable information in selecting the appropriate combinations of invertase and invertase inhibitor. Therefore, a better understanding of the interactions between specific invertase and invertase inhibitor alleles will be helpful for an intelligent manipulation of the cold-induced sweetening process of potato tubers.

An Unconstrained and Effective Approach of Script Identification for Online Bilingual Handwritten Text

A number of monolingual online handwriting recognition (OHR) systems were proposed by researchers for languages like English, Chinese, Japanese, Hindi, Punjabi, Tamil, Bangla, Arabic and many more. But very few researchers have worked on bilingual or multilingual OHR systems, as the challenging part of these systems is the identification of scripts during recognition of intermixed words. In this paper, a bilingual OHR system is proposed for the mixed text containing words, written by using two scripts Gurmukhi (for Punjabi language) and Roman (for English language). For converting handwritten intermixed text to digital text, two steps have been implemented: first, to identify the script of segmented handwritten word and second, to run the respective script recognition engine. Multilayered perceptron neural network classifier along with directional code feature has been implemented to identify the script of segmented handwritten word. Considerable results have been observed for the script identification process.