An Open Source Tesseract Based Optical Character Recognizer for Bengali Language

Optical Character Recognition (OCR) is the process of extracting text from an image. The main purpose of an OCR is to make editable documents from existing paper documents or image files. OCR primarily works in two phases; they are character and word detection. In case of more sophisticated approach, an OCR also works on sentence detection to preserve documents’ structures. In this paper, we would discuss the process of developing an OCR for Bengali language. Lots of efforts have been put on developing an OCR for Bengali. Though some OCRs have been developed, none of them is completely error free. For our thesis, we trained Tesseract OCR Engine to develop an OCR for Bengali language. Tesseract is currently the most accurate OCR engine. This engine was developed at HP labs and currently sponsored by Google. In Tesseract there are two option to training first one is Legacy Training and second is LSTM Training. We do both of them.

Who tweets climate change papers? Investigating publics of research through users’ descriptions

As social issues like climate change become increasingly salient, digital traces left by scholarly documents can be used to assess the resonance of scientific knowledge outside academia. Our research describes a method to assess the publics of research on Twitter by focusing on perceived users who shared highly tweeted climate change papers. It examines users through eight categories (academia, communication, political, professional, personal, organization, bots and publishers) associated to specific expressions in Twitter profile descriptions. Results indicate how diverse publics may be represented in the communication of scholarly documents on Twitter. Supplementing our word detection analysis with qualitative assessments of the results, we highlight how the presence of unique or multiple categorizations in textual Twitter descriptions provides evidence of the publics of research in specific contexts. The notion of perceived users allows to circumvent some issues about the construction of profiles through specific identity markers. Furthermore, the flexibility of our method provide means for research assessment that take into account the contextuality and plurality of publics involved on Twitter.

Learning and Retention of Novel Words in Musicians and Nonmusicians

Purpose The purpose of this study was to determine whether long-term musical training enhances the ability to perceive and learn new auditory information. Listeners with extensive musical experience were expected to detect, learn, and retain novel words more effectively than participants without musical training. Advantages of musical training were expected to be greater for words learned in multitalker babble compared to quiet. Method Participants consisted of 20 young adult musicians and 20 age-matched nonmusicians, all with normal hearing. In addition to completing word recognition and nonword detection tasks, each participant learned 10 novel words in a rapid word-learning paradigm. All tasks were completed in quiet and in multitalker babble. Next-day retention of the learned words was examined in isolation (recall) and in the context of continuous discourse (detection). Performance was compared across groups and listening conditions. Results Performance was significantly poorer in babble than in quiet on word recognition and nonword detection, but not on word learning, learned-word recall, or learned-word detection. No differences were observed between groups (musicians vs. nonmusicians) on any of the tasks. Conclusions For young normal-hearing adults, auditory experience resulting from long-term music training did not enhance their learning of new auditory information in either favorable (quiet) or unfavorable (babble) listening conditions. This suggests that the formation of semantic and musical representations in memory may be supported by the same underlying auditory processes, such that musical training is simply an extension of an auditory expertise that both musicians and nonmusicians possess.

Landslide Victories

Tensor decompositions are defined for deep learning networks and active filter designs for the class of problems of event detection and wake word detection filters, for wildlife and demographic vocalization and footstep census and for landslide detection. The problems are proven Z complete from previous work, in published literature. An estimate on the minimal number of samples required to predict demographic and wildlife census and reasonably predict landslides within given confidence intervals is presented using clustered, stratified sampling.The Shaktiman(™) is introduced as a USB form factor IP 68 system for integration into computing for IoT applications using SoC RF solutions similar to BOMU and TOMU using the Shakti Risc V processor developed at IIT Madras. The Thunderboard Sense 2 module is directly integrated to 3D printed mathematical art to create solar lanterns for hymnology and early bird warning systems, for data logging, bioluminosity, early bird warning systems for natural disasters like landslides and other weather disturbances, using integrated temperature, humidity and hall sensors.Cloud integration with Google Firebase is used for a FaaS framework to the use of tensor decomposition in defining the architecture of deep learning and procedural A.I for event detection from multi sensor fusion. A commercial product already available through Shapeways.com, designed by the author is to be enhanced to add event detection and wake word detection functions for PID systems and natural disaster monitoring and prediction infrastructure to add to the existing pioneering efforts by IIT Mandi.Keywords: Disaster Prediction, Landslide, Footstep detection, Air Pollution Monitoring, Solar Garden Lamps, Hymnology, Early Bird Warning Systems, Indigenous Whistle Languages

Codeword Detection, Focusing on Differences in Similar Words Between Two Corpora of Microblogs

Recently, the use of microblogs in drug trafficking has surged and become a social problem. A common method applied by cyber patrols to repress crimes, such as drug trafficking, involves searching for crime-related keywords. However, criminals who post crime-inducing messages maximally exploit “codewords” rather than keywords, such as enjo kosai, marijuana, and methamphetamine, to camouflage their criminal intentions. Research suggests that these codewords change once they gain popularity; thus, effective codeword detection requires significant effort to keep track of the latest codewords. In this study, we focused on the appearance of codewords and those likely to be included in incriminating posts to detect codewords with a high likelihood of inclusion in incriminating posts. We proposed new methods for detecting codewords based on differences in word usage and conducted experiments on concealed-word detection to evaluate the effectiveness of the method. The results showed that the proposed method could detect concealed words other than those in the initial list and to a better degree than the baseline methods. These findings demonstrated the ability of the proposed method to rapidly and automatically detect codewords that change over time and blog posts that instigate crimes, thereby potentially reducing the burden of continuous codeword surveillance.