Embedded Files
Research Journals
Research Journals
Real-Time Speech Emotion Analysis for Smart Home Assistants
Real-Time Speech Emotion Analysis for Smart Home Assistants
R. Chatterjee, S. Mazumdar, R. S. Sherratt, R. Halder, T. Maitra and D. Giri, "Real-Time Speech Emotion Analysis for Smart Home Assistants," in IEEE Transactions on Consumer Electronics, vol. 67, no. 1, pp. 68-76, Feb. 2021, doi: 10.1109/TCE.2021.3056421.
Abstract: Artificial Intelligence (AI) based Speech Emotion Recognition (SER) has been widely used in the consumer field for control of smart home personal assistants, with many such devices on the market. However, with the increase in computational power, connectivity, and the need to enable people to live in the home for longer though the use of technology, then smart home assistants that could detect human emotion will improve the communication between a user and the assistant enabling the assistant of offer more productive feedback. Thus, the aim of this work is to analyze emotional states in speech and propose a suitable method considering performance verses complexity for deployment in Consumer Electronics home products, and to present a practical live demonstration of the research. In this article, a comprehensive approach has been introduced for the human speech-based emotion analysis. The 1-D convolutional neural network (CNN) has been implemented to learn and classify the emotions associated with human speech. The paper has been implemented on the standard datasets (emotion classification) Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) and Toronto Emotional Speech Set database (TESS) (Young and Old). The proposed approach gives 90.48%, 95.79% and 94.47% classification accuracies in the aforementioned datasets. We conclude that the 1-D CNN classification models used in speaker-independent experiments are highly effective in the automatic prediction of emotion and are ideal for deployment in smart home assistants to detect emotion.
Conference Publications
Conference Publications
International Conference on Advances on Data Computing, Communication and Security (I3CS-2021)
International Conference on Advances on Data Computing, Communication and Security (I3CS-2021)
Deep Learning Approach for Motor-imagery Brain States Discrimination Problem
Deep Learning Approach for Motor-imagery Brain States Discrimination Problem
S. Mazumdar, R. Chatterjee
Abstract: Brain signals can be used to control robotic limbs for partial or fully paralyzed persons. Electroencephalography (EEG) is a widely used non-invasive brain signal recording technique. It is essential to process and understand the hidden patterns associated with a specific cognitive or motor task. Here, the focus is on motor-imagery (MI) EEG signal classification. There is a significant difference between machine learning and deep learning algorithms at the feature extraction phase. In this paper, a one-dimensional (1D) Convolutional Neural Network (CNN) has been proposed to interpret motor-imagery left-hand and right-hand movements. The proposed model has been compared with the existing SOTA techniques on the same BCI Competition II Dataset III. It outperforms the traditional machine learning models and achieves 91.43% classification accuracy.
URL: https://link.springer.com/book/10.1007/978-981-16-8403-6
Post-Graduation Thesis & Dissertations
Post-Graduation Thesis & Dissertations
Motor-Imagery Brain State Discrimination of Limbs
Motor-Imagery Brain State Discrimination of Limbs
S Mazumdar
Abstract: An electroencephalogram (EEG) is a widely used painless, non-invasive procedure to collect motor-imagery signals from the brain. Moreover EEG has an extensive contribution on determining different limb movements. Immobilized persons can further use motor-imagery signals to move robotic limbs and this process needs to understand the hindrance pattern associated with a specific cognitive or motor-task. In this project a Daubechius Wavelet based feature extraction followed by a one dimensional Convolutional Neural Network architecture has been proposed to distinguish motor-imagery (MI) left-hand & right-hand movement. The model has been built based on BCI Competition IV Dataset II-B and compared with existing SOTA techniques on the same dataset. It outperforms the traditional models and archives a state of the art 97.2% classification accuracy.
Graduation Thesis & Dissertations
Graduation Thesis & Dissertations
An Overview of Receivables Analytics for Business Intelligence
An Overview of Receivables Analytics for Business Intelligence
S Mazumdar
Abstract: HighRadius works on the account receivables side of the transaction process between the client and buyer. High Radius empowers corporations to modernize receivables in order to lower Days Sales Outstanding (DSO), minimize write-offs, and reduce operating expenses. High Radius products like Receivables Cloud and Payments Cloud helps corporations by providing filtered data and proper information about the payments, but basically analysis of the data manually by the analysts require a lot of time. Therefore, in this project we tried to empower Artificial Intelligence Rivana to solve the problems using classification algorithms. HighRadius Deductions Cloud enables a proactive deduction management operation. The solution streamlines processing, shortens resolution cycle time, reduces processing costs, and increases recovery rates on invalid deductions. But Deduction Management teams and analysts process hundreds of thousands of deductions every year. However, even if a deduction is valid, it still requires a set of manual and time-consuming tasks to be executed before an analyst is able to determine its validity. With more than half of all deductions being valid, this means that credit and A/R teams lose productivity that could have been spent on resolving and collecting on invalid deductions. HighRadius Rivana allows deduction analyst to focus on resolving disputes which are more likely to be invalid. Employing classification algorithms, Rivana is capable of identifying deductions which have a very high probability of being valid and is able to automatically resolve them or move them or de-prioritize them on the analyst’s worklist. The main objective of the project is to classify the raised deductions to be either valid or invalid by leveraging machine learning algorithms on the historical data.
Linear Quadratic Estimation using Kalman Filter and Application
Linear Quadratic Estimation using Kalman Filter and Application
S Mazumdar
Abstract: For estimations in statistics and control theory, Linear Quadratic Estimation (LQE) plays a different role due to the use of optimal algorithms. Kalman Filtering can be used to predict parameters of interest, such as location, speed, the direction in the presence of noisy measurements. It is a widely used filtration technology for Signal Processing. The very first use of Kalman Filter was during “Project Apollo” to estimate trajectories of the manned spacecraft to the moon and back. Prediction of measurements using several machine learning and statistical algorithms are heavyweight and provide results depending on provided parameters. It’s complex to design self-adaptive nature in different statistical methods, whereas Kalman Filtering works completely on its error differentiating mechanism. Which is more lightweight than a few ML dependant procedures. Being linear, it’s fast and effective in deflection path detection. State Observation method helps to estimate something, which can not be seen or measured directly. From very basic adaptive filtration technologies, state observation has been used to minimize the error of the mathematical estimators and practical values. LQE compares the available methods and calculates MSE which is used to minimize the expected system error.
Enhanced Analytic System for Smart University Assistance
Enhanced Analytic System for Smart University Assistance
N Brahmachari, R Bordoloi, R Das, S Mazumdar, S Bakshi
Abstract: Due to the increase in the number of enrollments, efficiently managing the overall system, without useless traffic has become a laborious job. "BeFriend" helps the higher up authority to organize the complete process efficiently. It is a user-friendly UX rich system which prioritizes and arrange the available tasks accordingly.
The complete system is based on a Client-Server three-tier architecture, where the server is separated from user access and based on a cloud system. The system also brings an analytical comparison with the details collected from students. A hostel allocation methodology has been integrated into the system to allocate a room of their choice. Two standalone machine learning - ensemble-based analysis is used in the integration to help the students for suggestion and selection of stream and well-structured career path. To avoid accuracy conflict, our propose architecture shows a set of fuzzy data along with relative choices. A synthesized dataset is developed for the stream selection functionality and a benchmark AMEO-2015 dataset for career guidance. We achieved an accuracy of 98% and 66% in the validation sets.
Additionally, a highly secure admin portal is provided with the system which used AR to verify admin marker. An alumni chat is there to connect alumnus through the system. For present students, a dashboard and a daily schedule are generated automatically in the student home. Besides, a task management system is provided with the home to manage the tasks daily.
Scan for an overview
Page updated
Google Sites
Report abuse