HiEn: Speak the Switch
DOI:
https://doi.org/10.63503/acset.978-81-995593-9-4.56Keywords:
Machine Translation, OCR, Speech Recognition, Ancient Languages, Brahmi Script, Multimodal System, Neural Machine TranslationAbstract
Linguistic barriers remain a major challenge in communication, education, and cultural preservation, particularly for ancient and low-resource languages. While modern neural machine translation systems perform efficiently for widely used languages, they lack sufficient support for historical scripts such as Brahmi and classical languages due to limited datasets. This paper presents HiEn: Speak the Switch, a multimodal multilingual conversion system that integrates text, image, and voice-based translation within a unified platform. The system combines Optical Character Recognition, speech recognition, text-to-speech synthesis, and machine translation APIs within a Python-based graphical interface. Additionally, a custom character-mapping module enables deterministic conversion of the Brahmi script into modern equivalents. The system demonstrates reliable multilingual translation and satisfactory OCR performance. It enhances accessibility, simplifies translation workflows, and contributes to the digital preservation of ancient scripts. The architecture is scalable and can be extended using advanced transformer-based models in future implementations.
References
[1] M. Lupaşcu, Large multimodal models of low-resource languages
[2] Languages, Journal of Artificial Intelligence Research, vol. 76, pp. 145178, 2026.
[3] N. Sethiya et al., “Indic-ST: A large-scale multilingual corpus of speech translation, low-resource speech translation, ACM Trans. Asian Low-Resour. Lang. Inf. Process., vol. 24, no. 3, pp. 1– 21, 2025.
[4] N. Kadyrbek, A. Issa, and M. Al-Khalifa, Development of small-scale language models on low- resource languages, Big Data Cogn. Comput., vol. 9, no. 5, pp. 137–152, 2025.
[5] J. Iranzo-Sánchez et al., Speech translation to multilingual medical education Artif. Intell. Med., vol. 145,
[6] pp. 102–118, 2025.
[7] S. Roy, “Multi-language translation system based on AI: Architecture/performance analysis of the system-on-chip, Research Preprint, 2025.
[8] P. Kocmi and O. Bojar, Trivial transfer learning to low-resource neural machine translator, in Proc. WMT, 2024, pp. 120313.
[9] A. Pope, D. Chen, and L. Smith, BLEU and beyond: Evaluation measures of neural machine translation, Comput—Linguistics, vol. 50, no. 2, pp. 355–372, 2024.
[10] S. Klejch et al, OCR error correction of low-resource scripts, in Proc. ICDAR, 2024, pp. 98105.
[11] D. Hughes and M. Brown, “Transformer-based OCR to digitise documents, Pattern Recognit. Lett., vol. 172, pp. 55–63, 2023.
[12] P. Winata et al., Multilingual self-attention to low-resource speech recognition, in Proc. ACL,
[13] pp. 231242.
[14] A. Conneau et al., “Unsupervised cross-lingual representation learning at scale, Trans. Assoc. Comput. Linguistics, vol. 10, pp. 125–141, 2022.
[15] Y. Liu et al., XLM-R: Cross-lingual language model pretraining, in Proc. ACL, 2022, pp. 1 15.
[16] A. Kahn et al., Neural machine translation evaluation methods revisited, in Proc. EMNLP in 2021, pp. 678 689.
[17] A. Graves et al., Speech recognition using deep recurrent neural networks, in Proc. ICASSP, 2013, pp. 6645 6649.
[18] J. Devlin et al., BERT: Pre-training of deep bidirectional transformers, in Proc. NAACL-HLT, 2019,
[19] pp. 41714186.
[20] A. Vaswani et al., Attention is all you need, in Proc. NeurIPS, 2017, pp. 59986008.
[21] D. Bahdanau, K. Cho, and Y. Bengio, “Neural machine translation through jointly learning to align and translate, in Proc. ICLR, 2015.
[22] I. Sutskever, O. Vinyals, and Q. Le, “Sequence to sequence learning using neural networks, in Proc. NeurIPS, 2014, pp. 31043112.
[23] R. Smith, “An overview of the Tesseract OCR engine, in Proc. ICDAR, 2007, pp. 629633.
[24] P. Koehn, Statistical Machine Translation. Cambridge, U.K.: Camb. Univ. Press, 2009.
