Publications

2026

1. NLPJ

   Kingfisher: A hybrid LLM-augmented Bangla text normalization for enhanced text-to-speech

   Rajan Saha Raju, Arif Ahmad, and Mohammad Shahidur Rahman

   Natural Language Processing Journal, 2026

   Abs DOI Bib Code

   Text normalization (TN), the process of converting non-standard words into their spoken equivalents, is a fundamental pre-processing step for text-to-speech (TTS) systems. While substantial progress has been made in TN for well-resourced languages, low-resource languages such as Bangla have received limited attention. We present Kingfisher, a three-stage hybrid framework combining LLM-based tokenization and semiotic class annotation, lexicon-driven context-aware verbalization, and error correction to build an accurate Bangla text normalizer. Experimental evaluations across diverse Bangla texts demonstrate that Kingfisher achieves superior performance, with an overall accuracy of 96% (confidence interval 95%–97%), significantly outperforming the only publicly available Bangla text normalizer, Sparrowhawk. To support further research, we release the Bangla text normalization dataset and make the source code of the text normalization system publicly available11https://github.com/Rajan-sust/BanglaTextNorm, offering a substantial contribution to the Bangla speech technology community.

   @article{kingfisher2026,
     title = {Kingfisher: A hybrid LLM-augmented Bangla text normalization for enhanced text-to-speech},
     journal = {Natural Language Processing Journal},
     volume = {14},
     pages = {100201},
     year = {2026},
     issn = {2949-7191},
     doi = {https://doi.org/10.1016/j.nlp.2026.100201},
     url = {https://www.sciencedirect.com/science/article/pii/S2949719126000051},
     author = {Raju, Rajan Saha and Ahmad, Arif and Rahman, Mohammad Shahidur},
     keywords = {Text normalization, Text-to-speech (TTS), Bangla, Large language model (LLM)},
   }

2025

1. bioRxiv

   ProtSEC: Ultrafast Protein Sequence Embedding in Complex Space Using Fast Fourier Transform

   Rajan Saha Raju and Rashedul Islam

   bioRxiv, 2025

   Abs DOI Bib Code

   Among the various approaches for representing protein sequences as vectors, embeddings derived from protein language models (PLMs) have been empirically shown to enhance accuracy in downstream bioinformatics tasks. However, the substantial computational demands of PLMs, both during training and inference, pose significant challenges. We introduce ProtSEC (Protein Sequence Embedding in Complex Space), a novel approach that encodes each amino acid as a unique complex number derived from the BLOSUM62 substitution matrix. By modeling protein sequences as complex signals and applying the Fast Fourier Transform (FFT), ProtSEC generates embeddings in the complex space. Unlike PLMs, ProtSEC requires no pre-training on large protein sequence datasets and operates independently of any pre-trained models. Our benchmarking demonstrate that ProtSEC achieves a 20,000-fold reduction in runtime and an 85-fold improvement in memory efficiency compared to popular PLMs (e.g., esm2_3B, esm2_35M, prot_t5, prot_bert). Depending on the task, ProtSEC demonstrates either superior or comparable accuracy to PLMs in sequence similarity search, sequence classification and phylogenetic tree reconstruction. ProtSEC provides fast and accurate protein sequence embeddings in complex numbers, facilitating efficient integration into diverse downstream bioinformatics workflows. ProtSEC is available at https://github.com/omics-lab/ProtSEC.Competing Interest StatementThe authors have declared no competing interest.

   @article{protsec2025,
     author = {Raju, Rajan Saha and Islam, Rashedul},
     title = {ProtSEC: Ultrafast Protein Sequence Embedding in Complex Space Using Fast Fourier Transform},
     elocation-id = {2025.08.17.670693},
     year = {2025},
     doi = {10.1101/2025.08.17.670693},
     publisher = {Cold Spring Harbor Laboratory},
     url = {https://www.biorxiv.org/content/early/2025/10/09/2025.08.17.670693},
     journal = {bioRxiv}
   }

2022

1. Processes

   A Reinforcement-Learning-Based Model for Resilient Load Balancing in Hyperledger Fabric

   Reem Alotaibi, Madini Alassafi, Md. Saiful Islam Bhuiyan, and 2 more authors

   Processes, 2022

   Abs DOI

   Blockchain with its numerous advantages is often considered a foundational technology with the potential to revolutionize a wide range of application domains, including enterprise applications. These enterprise applications must meet several important criteria, including scalability, performance, and privacy. Enterprise blockchain applications are frequently constructed on private blockchain platforms to satisfy these criteria. Hyperledger Fabric is one of the most popular platforms within this domain. In any privacy blockchain system, including Fabric, every organisation needs to utilise a peer node (or peer nodes) to connect to the blockchain platform. Due to the ever-increasing size of blockchain and the need to support a large user base, the monitoring and the management of different resources of such peer nodes can be crucial for a successful deployment of such blockchain platforms. Unfortunately, little attention has been paid to this issue. In this work, we propose the first-ever solution to this significant problem by proposing an intelligent control system based on reinforcement learning for distributing the resources of Hyperledger Fabric. We present the architecture, discuss the protocol flows, outline the data collection methods, analyse the results and consider the potential applications of the proposed approach.

2. Genomics

   VirusTaxo: Taxonomic classification of viruses from the genome sequence using k-mer enrichment

   Rajan Saha Raju, Abdullah Al Nahid, Preonath Chondrow Dev, and 1 more author

   Genomics, 2022

   Abs DOI Bib Code

   Classification of viruses into their taxonomic ranks (e.g., order, family, and genus) provides a framework to organize an abundant population of viruses. Next-generation metagenomic sequencing technologies lead to a rapid increase in generating sequencing data of viruses which require bioinformatics tools to analyze the taxonomy. Many metagenomic taxonomy classifiers have been developed to study microbiomes, but it is particularly challenging to assign the taxonomy of diverse virus sequences and there is a growing need for dedicated methods to be developed that are optimized to classify virus sequences into their taxa. For taxonomic classification of viruses from metagenomic sequences, we developed VirusTaxo using diverse (e.g., 402 DNA and 280 RNA) genera of viruses. VirusTaxo has an average accuracy of 93% at genus level prediction in DNA and RNA viruses. VirusTaxo outperformed existing taxonomic classifiers of viruses where it assigned taxonomy of a larger fraction of metagenomic contigs compared to other methods. Benchmarking of VirusTaxo on a collection of SARS-CoV-2 sequencing libraries and metavirome datasets suggests that VirusTaxo can characterize virus taxonomy from highly diverse contigs and provide a reliable decision on the taxonomy of viruses.

   @article{RAJU2022110414,
     title = {VirusTaxo: Taxonomic classification of viruses from the genome sequence using k-mer enrichment},
     journal = {Genomics},
     volume = {114},
     number = {4},
     pages = {110414},
     year = {2022},
     issn = {0888-7543},
     doi = {https://doi.org/10.1016/j.ygeno.2022.110414},
     url = {https://www.sciencedirect.com/science/article/pii/S0888754322001598},
     author = {Raju, Rajan Saha and {Al Nahid}, Abdullah and {Chondrow Dev}, Preonath and Islam, Rashedul},
     keywords = {Virus, Taxonomy, Hierarchical classification, k-mer, Genome}
   }

2021

1. BiB

   Choice of assemblers has a critical impact on de novo assembly of SARS-CoV-2 genome and characterizing variants

   Rashedul Islam, Rajan Saha Raju, Nazia Tasnim, and 4 more authors

   Briefings in Bioinformatics, Sep 2021

   Abs DOI Bib

   Coronavirus Disease 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has become a global pandemic following its initial emergence in China. SARS-CoV-2 has a positive-sense single-stranded RNA virus genome of around 30Kb. Using next-generation sequencing technologies, a large number of SARS-CoV-2 genomes are being sequenced at an unprecedented rate and being deposited in public repositories. For the de novo assembly of the SARS-CoV-2 genomes, a myriad of assemblers is being used, although their impact on the assembly quality has not been characterized for this virus. In this study, we aim to understand the variabilities on assembly qualities due to the choice of the assemblers.We performed 6648 de novo assemblies of 416 SARS-CoV-2 samples using eight different assemblers with different k-mer lengths. We used Illumina paired-end sequencing reads and compared the assembly quality of those assemblers. We showed that the choice of assembler plays a significant role in reconstructing the SARS-CoV-2 genome. Two metagenomic assemblers, e.g. MEGAHIT and metaSPAdes, performed better compared with others in most of the assembly quality metrics including, recovery of a larger fraction of the genome, constructing larger contigs and higher N50, NA50 values, etc. We showed that at least 09% (259/2873) of the variants present in the assemblies between MEGAHIT and metaSPAdes are unique to one of the assembly methods.Our analyses indicate the critical role of assembly methods for assembling SARS-CoV-2 genome using short reads and their impact on variant characterization. This study could help guide future studies to determine the best-suited assembler for the de novo assembly of virus genomes.

   @article{10.1093/bib/bbab102,
     author = {Islam, Rashedul and Raju, Rajan Saha and Tasnim, Nazia and Shihab, Istiak Hossain and Bhuiyan, Maruf Ahmed and Araf, Yusha and Islam, Tofazzal},
     title = {Choice of assemblers has a critical impact on de novo assembly of SARS-CoV-2 genome and characterizing variants},
     journal = {Briefings in Bioinformatics},
     volume = {22},
     number = {5},
     pages = {bbab102},
     year = {2021},
     month = sep,
     issn = {1477-4054},
     doi = {10.1093/bib/bbab102},
     url = {https://doi.org/10.1093/bib/bbab102},
   }

2. ICSCT

   End-to-End Bangla Speech Synthesis

   Prithwiraj Bhattacharjee, Rajan Saha Raju, Arif Ahmad, and 1 more author

   In 2021 International Conference on Science & Contemporary Technologies (ICSCT), 2021

   Abs DOI Bib

   Text-to-Speech (TTS) system is a system where speech is synthesized from a given text following any particular approach. Concatenative synthesis, Hidden Markov Model (HMM) based synthesis, Deep Learning (DL) based synthesis with multiple building blocks, etc. are the main approaches for implementing a TTS system. Here, we are presenting our deep learning-based end-to-end Bangla speech synthesis system. It has been implemented with minimal human annotation using only 3 major components (Encoder, Decoder, Post-processing net including waveform synthesis). It does not require any frontend preprocessor and Grapheme-to-Phoneme (G2P) converter. Our model has been trained with phonetically balanced 20 hours of single speaker speech data. It has obtained a 3.79 Mean Opinion Score (MOS) on a scale of 5.0 as subjective evaluation and a 0.77 Perceptual Evaluation of Speech Quality(PESQ) score on a scale of [-0.5,4.5] as objective evaluation. It is outperforming all existing non-commercial state-of-the-art Bangla TTS systems based on naturalness.

   @inproceedings{9642672,
     author = {Bhattacharjee, Prithwiraj and Raju, Rajan Saha and Ahmad, Arif and Rahman, M. Shahidur},
     booktitle = {2021 International Conference on Science & Contemporary Technologies (ICSCT)},
     title = {End-to-End Bangla Speech Synthesis},
     year = {2021},
     volume = {},
     number = {},
     pages = {1-6},
     keywords = {Deep learning;Automation;Annotations;Hidden Markov models;Data models;Decoding;Speech synthesis;Text-to-Speech;End-to-End;MOS;PESQ;Tacotron;Griffin-lim},
     doi = {10.1109/ICSCT53883.2021.9642672}
   }

2019

1. ICBSLP

   A Bangla Text-to-Speech System using Deep Neural Networks

   Rajan Saha Raju, Prithwiraj Bhattacharjee, Arif Ahmad, and 1 more author

   In 2019 International Conference on Bangla Speech and Language Processing (ICBSLP), 2019

   Abs DOI Bib

   We present a Deep Neural Network (DNN) based statistical parametric Text-to-Speech (TTS) system for Bangla (also known as Bengali). A first step in building a DNN-based TTS system is having large speech data. Since good speech dataset for Bangla TTS is not available publicly, we created our own dataset for our system. We prepared a phonetically rich studio-quality speech database containing more than 40 hours of speech. The database consists of 12,500 utterances. We also prepared a pronunciation dictionary (lexicon) of 1,35,000 words for front-end text processing, which, to our knowledge, is the largest lexicon for Bangla. Our system extracts linguistic features from input text. Then it uses deep neural networks for mapping these linguistic features to acoustic features. We developed two TTS voices using our dataset - one male and one female voice. Both objective and subjective evaluation tests show that our system performs significantly better than the traditional Bangla TTS systems and is comparable to the commercially available best Bangla TTS system.

   @inproceedings{9084052,
     author = {Raju, Rajan Saha and Bhattacharjee, Prithwiraj and Ahmad, Arif and Rahman, Mohammad Shahidur},
     booktitle = {2019 International Conference on Bangla Speech and Language Processing (ICBSLP)},
     title = {A Bangla Text-to-Speech System using Deep Neural Networks},
     year = {2019},
     volume = {},
     number = {},
     pages = {1-5},
     keywords = {Dictionaries;Databases;Scalability;Buildings;Lexicon;Artificial neural networks;Linguistics;Feature extraction;Acoustics;Text to speech;spss;dnn;bangla speech corpus;lexicon;open source},
     doi = {10.1109/ICBSLP47725.2019.202055}
   }