Rajesh Kumar Aggarwal
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Rajesh Kumar Aggarwal has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Mayank Dave
National Institute of Technology
5 shared publications - Ankit Kumar
Department of Computer Engineering, National institute of technology, Kurukshetra, India
3 shared publications - Vijay Verma
National Institute of Technology
3 shared publications Vishal Passricha
National Institute of Technology
3 shared publications- Vishal Passricha
Central University of Haryana
2 shared publications - Anisha Datta
National Institute of Technology
1 shared publication
Latest Publications
- 1 August 2025
- Published by Springer Nature
- p. 503-512
Human Activities Recognition Using Transformer Encoder-Only Model
- 1 August 2025
- Published by Springer Nature
- p. 503-512
Human Activities Recognition Using Transformer Encoder-Only Model
- 8 April 2021
- Published by Springer Nature in Journal of Reliable Intelligent Environments
- Vol. 8 (2) , 117-132
An exploration of semi-supervised and language-adversarial transfer learning using hybrid acoustic model for hindi speech recognition
- 8 April 2021
- Published by Springer Nature in Journal of Reliable Intelligent Environments
- Vol. 8 (2) , 117-132
An exploration of semi-supervised and language-adversarial transfer learning using hybrid acoustic model for hindi speech recognition
- Open Access
- 3 November 2020
- Published by AGH University Press in Computer Science
- Vol. 21 (4) , 397-417
A hybrid CNN-LiGRU acoustic modeling using raw waveform sincnet for Hindi ASR
Deep Neural Network (DNN) is currently playing the most vital role in Automatic Speech Recognition (ASR). Convolution Neural Network (CNN) and Recurrent Neural Network (RNN) are the advanced versions of DNN. CNN and RNN are right to deal with spatial and temporal properties of the speech signal, respectively, and both properties have a higher impact on accuracy. In today’s scenario, many acoustic modeling techniques often switches due to the battle of CNNs and RNNs. In the last few years, CNN, with raw speech signal, shows their superiority over precomputed acoustic features. Recently, a novel first convolution layer named SincNet was proposed to produce the interpretable filters with better accuracy. In this work, we proposed a hybrid SincNet-CNN-RNN architecture with low computation cost and high accuracy. Different configurations of the hybrid model were extensively examined to achieve this goal. All experiments were performed on the Hindi speech dataset.
- 3 November 2020
- Published by AGH University Press in Computer Science
- Vol. 21 (4) , 397-417
Open AccessA hybrid CNN-LiGRU acoustic modeling using raw waveform sincnet for Hindi ASR
Deep Neural Network (DNN) is currently playing the most vital role in Automatic Speech Recognition (ASR). Convolution Neural Network (CNN) and Recurrent Neural Network (RNN) are the advanced versions of DNN. CNN and RNN are right to deal with spatial and temporal properties of the speech signal, respectively, and both properties have a higher impact on accuracy. In today’s scenario, many acoustic modeling techniques often switches due to the battle of CNNs and RNNs. In the last few years, CNN, with raw speech signal, shows their superiority over precomputed acoustic features. Recently, a novel first convolution layer named SincNet was proposed to produce the interpretable filters with better accuracy. In this work, we proposed a hybrid SincNet-CNN-RNN architecture with low computation cost and high accuracy. Different configurations of the hybrid model were extensively examined to achieve this goal. All experiments were performed on the Hindi speech dataset.
- Open Access
- 28 July 2020
- Published by Walter de Gruyter GmbH in Journal of Intelligent Systems
- Vol. 30 (1) , 165-179
Discriminatively trained continuous Hindi speech recognition using integrated acoustic features and recurrent neural network language modeling
This paper implements the continuous Hindi Automatic Speech Recognition (ASR) system using the proposed integrated features vector with Recurrent Neural Network (RNN) based Language Modeling (LM). The proposed system also implements the speaker adaptation using Maximum-Likelihood Linear Regression (MLLR) and Constrained Maximum likelihood Linear Regression (C-MLLR). This system is discriminatively trained by Maximum Mutual Information (MMI) and Minimum Phone Error (MPE) techniques with 256 Gaussian mixture per Hidden Markov Model(HMM) state. The training of the baseline system has been done using a phonetically rich Hindi dataset. The results show that discriminative training enhances the baseline system performance by up to 3%. Further improvement of ~7% has been recorded by applying RNN LM. The proposed Hindi ASR system shows significant performance improvement over other current state-of-the-art techniques.
- 28 July 2020
- Published by Walter de Gruyter GmbH in Journal of Intelligent Systems
- Vol. 30 (1) , 165-179
Open AccessDiscriminatively trained continuous Hindi speech recognition using integrated acoustic features and recurrent neural network language modeling
This paper implements the continuous Hindi Automatic Speech Recognition (ASR) system using the proposed integrated features vector with Recurrent Neural Network (RNN) based Language Modeling (LM). The proposed system also implements the speaker adaptation using Maximum-Likelihood Linear Regression (MLLR) and Constrained Maximum likelihood Linear Regression (C-MLLR). This system is discriminatively trained by Maximum Mutual Information (MMI) and Minimum Phone Error (MPE) techniques with 256 Gaussian mixture per Hidden Markov Model(HMM) state. The training of the baseline system has been done using a phonetically rich Hindi dataset. The results show that discriminative training enhances the baseline system performance by up to 3%. Further improvement of ~7% has been recorded by applying RNN LM. The proposed Hindi ASR system shows significant performance improvement over other current state-of-the-art techniques.
- 24 October 2019
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 22 (4) , 1123-1133
PSO-based optimized CNN for Hindi ASR
- 24 October 2019
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 22 (4) , 1123-1133
PSO-based optimized CNN for Hindi ASR
- Open Access
- 8 June 2019
- Published by MECS Publisher in International Journal of Information Technology and Computer Science
- Vol. 11 (6) , 37-49
A New Similarity Measure Based on Simple Matching Coefficient for Improving the Accuracy of Collaborative Recommendations
Recommender Systems (RSs) are essential tools of an e-commerce portal in making intelligent decisions for an individual to obtain product recommendations. Neighborhood-based approaches are traditional techniques for collaborative recommendations and are very popular due to their simplicity and efficiency. Neighborhood-based recommender systems use numerous kinds of similarity measures for finding similar users or items. However, the existing similarity measures function only on common ratings between a pair of users (i.e. ignore the uncommon ratings) thus do not utilize all ratings made by a pair of users. Furthermore, the existing similarity measures may either provide inadequate results in many situations that frequently occur in sparse data or involve very complex calculations. Therefore, there is a compelling need to define a similarity measure that can deal with such issues. This research proposes a new similarity measure for defining the similarities between users or items by using the rating data available in the user-item matrix. Firstly, we describe a way for applying the simple matching coefficient (SMC) to the common ratings between users or items. Secondly, the structural information between the rating vectors is exploited using the Jaccard index. Finally, these two factors are leveraged to define the proposed similarity measure for better recommendation accuracy. For evaluating the effectiveness of the proposed method, several experiments have been performed using standardized benchmark datasets (MovieLens-1M, 10M, and 20M). Results obtained demonstrate that the proposed method provides better predictive accuracy (in terms of MAE and RMSE) along with improved classification accuracy (in terms of precision-recall).
- 8 June 2019
- Published by MECS Publisher in International Journal of Information Technology and Computer Science
- Vol. 11 (6) , 37-49
Open AccessA New Similarity Measure Based on Simple Matching Coefficient for Improving the Accuracy of Collaborative Recommendations
Recommender Systems (RSs) are essential tools of an e-commerce portal in making intelligent decisions for an individual to obtain product recommendations. Neighborhood-based approaches are traditional techniques for collaborative recommendations and are very popular due to their simplicity and efficiency. Neighborhood-based recommender systems use numerous kinds of similarity measures for finding similar users or items. However, the existing similarity measures function only on common ratings between a pair of users (i.e. ignore the uncommon ratings) thus do not utilize all ratings made by a pair of users. Furthermore, the existing similarity measures may either provide inadequate results in many situations that frequently occur in sparse data or involve very complex calculations. Therefore, there is a compelling need to define a similarity measure that can deal with such issues. This research proposes a new similarity measure for defining the similarities between users or items by using the rating data available in the user-item matrix. Firstly, we describe a way for applying the simple matching coefficient (SMC) to the common ratings between users or items. Secondly, the structural information between the rating vectors is exploited using the Jaccard index. Finally, these two factors are leveraged to define the proposed similarity measure for better recommendation accuracy. For evaluating the effectiveness of the proposed method, several experiments have been performed using standardized benchmark datasets (MovieLens-1M, 10M, and 20M). Results obtained demonstrate that the proposed method provides better predictive accuracy (in terms of MAE and RMSE) along with improved classification accuracy (in terms of precision-recall).
- 24 May 2019
- Published by Springer Nature in Journal of Ambient Intelligence and Humanized Computing
- Vol. 11 (2) , 675-691
A comparative analysis of pooling strategies for convolutional neural network based Hindi ASR
- 24 May 2019
- Published by Springer Nature in Journal of Ambient Intelligence and Humanized Computing
- Vol. 11 (2) , 675-691
A comparative analysis of pooling strategies for convolutional neural network based Hindi ASR
- Open Access
- 8 May 2019
- Published by MECS Publisher in International Journal of Modern Education and Computer Science
- Vol. 11 (5) , 41-53
Accuracy Assessment of Similarity Measures in Collaborative Recommendations Using CF4J Framework
There are various libraries that facilitate the design and development of recommender systems (RSs) research in both the academia and industry. Different libraries provide a different set of functionalities based on their foundational design principles. When new algorithms are proposed, researchers need to compare these against prior algorithms considering many challenges such as reproducibility of results, evaluation metrics, test harnesses, etc. Although many open source RS libraries exist to carry out research experiments and provide a varying degree of features such as extensibility, performance, scalability, flexibility, etc. To that end, this paper describes a recently introduced open-source RS library, Collaborative Filtering for Java (CF4J), which is specially designed for collaborative recommendations. Firstly, the brief internals of the CF4J framework are explained and it has been compared with other related libraries such as LibRec, LensKit, and Apache Mahout based on the recommendation approaches and evaluation tools. Secondly, we have summarized all the state-of-art similarity measures provided by the CF4J library. Finally, in order to determine the accuracy of these similarity measures, several experiments have been conducted using standardized benchmark datasets such as MovieLens-1M, MovieLens-10M, and MovieLens-20M. Empirically obtained results demonstrate that the Jaccard-Mean Squared Difference (JMSD) similarity measure provides better recommendation accuracy among all similarity measures.
- 8 May 2019
- Published by MECS Publisher in International Journal of Modern Education and Computer Science
- Vol. 11 (5) , 41-53
Open AccessAccuracy Assessment of Similarity Measures in Collaborative Recommendations Using CF4J Framework
There are various libraries that facilitate the design and development of recommender systems (RSs) research in both the academia and industry. Different libraries provide a different set of functionalities based on their foundational design principles. When new algorithms are proposed, researchers need to compare these against prior algorithms considering many challenges such as reproducibility of results, evaluation metrics, test harnesses, etc. Although many open source RS libraries exist to carry out research experiments and provide a varying degree of features such as extensibility, performance, scalability, flexibility, etc. To that end, this paper describes a recently introduced open-source RS library, Collaborative Filtering for Java (CF4J), which is specially designed for collaborative recommendations. Firstly, the brief internals of the CF4J framework are explained and it has been compared with other related libraries such as LibRec, LensKit, and Apache Mahout based on the recommendation approaches and evaluation tools. Secondly, we have summarized all the state-of-art similarity measures provided by the CF4J library. Finally, in order to determine the accuracy of these similarity measures, several experiments have been conducted using standardized benchmark datasets such as MovieLens-1M, MovieLens-10M, and MovieLens-20M. Empirically obtained results demonstrate that the Jaccard-Mean Squared Difference (JMSD) similarity measure provides better recommendation accuracy among all similarity measures.
- Open Access
- 5 March 2019
- Published by Walter de Gruyter GmbH in Journal of Intelligent Systems
- Vol. 29 (1) , 1261-1274
A Hybrid of Deep CNN and Bidirectional LSTM for Automatic Speech Recognition
Deep neural networks (DNNs) have been playing a significant role in acoustic modeling. Convolutional neural networks (CNNs) are the advanced version of DNNs that achieve 4–12% relative gain in the word error rate (WER) over DNNs. Existence of spectral variations and local correlations in speech signal makes CNNs more capable of speech recognition. Recently, it has been demonstrated that bidirectional long short-term memory (BLSTM) produces higher recognition rate in acoustic modeling because they are adequate to reinforce higher-level representations of acoustic data. Spatial and temporal properties of the speech signal are essential for high recognition rate, so the concept of combining two different networks came into mind. In this paper, a hybrid architecture of CNN-BLSTM is proposed to appropriately use these properties and to improve the continuous speech recognition task. Further, we explore different methods like weight sharing, the appropriate number of hidden units, and ideal pooling strategy for CNN to achieve a high recognition rate. Specifically, the focus is also on how many BLSTM layers are effective. This paper also attempts to overcome another shortcoming of CNN, i.e. speaker-adapted features, which are not possible to be directly modeled in CNN. Next, various non-linearities with or without dropout are analyzed for speech tasks. Experiments indicate that proposed hybrid architecture with speaker-adapted features and maxout non-linearity with dropout idea shows 5.8% and 10% relative decrease in WER over the CNN and DNN systems, respectively.
- 5 March 2019
- Published by Walter de Gruyter GmbH in Journal of Intelligent Systems
- Vol. 29 (1) , 1261-1274
Open AccessA Hybrid of Deep CNN and Bidirectional LSTM for Automatic Speech Recognition
Deep neural networks (DNNs) have been playing a significant role in acoustic modeling. Convolutional neural networks (CNNs) are the advanced version of DNNs that achieve 4–12% relative gain in the word error rate (WER) over DNNs. Existence of spectral variations and local correlations in speech signal makes CNNs more capable of speech recognition. Recently, it has been demonstrated that bidirectional long short-term memory (BLSTM) produces higher recognition rate in acoustic modeling because they are adequate to reinforce higher-level representations of acoustic data. Spatial and temporal properties of the speech signal are essential for high recognition rate, so the concept of combining two different networks came into mind. In this paper, a hybrid architecture of CNN-BLSTM is proposed to appropriately use these properties and to improve the continuous speech recognition task. Further, we explore different methods like weight sharing, the appropriate number of hidden units, and ideal pooling strategy for CNN to achieve a high recognition rate. Specifically, the focus is also on how many BLSTM layers are effective. This paper also attempts to overcome another shortcoming of CNN, i.e. speaker-adapted features, which are not possible to be directly modeled in CNN. Next, various non-linearities with or without dropout are analyzed for speech tasks. Experiments indicate that proposed hybrid architecture with speaker-adapted features and maxout non-linearity with dropout idea shows 5.8% and 10% relative decrease in WER over the CNN and DNN systems, respectively.