Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{201384,
        author = {Shravani Khamkar and Sneha Mishra and Huda Qazi Syed},
        title = {AnuvadX: A Translator for All 22 Scheduled Indian Languages},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {12},
        pages = {4255-4262},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=201384},
        abstract = {India’s linguistic diversity, which encompasses 22 constitutionally recognized languages, presents persistent challenges for inclusive digital communication. Existing translation systems often support limited languages or modalities of single interaction, restricting the applicability of multilingual systems in the real-world. This paper presents AnuvadX, a production-grade full-stack multilingual translation platform designed to support all 22 scheduled Indian languages across multiple input and output modalities. The system delivers text-to-text translation, PDF document translation, real-time voice recognition, and neural text-to-speech synthesis within a unified architecture. The frontend is built on Next.js with React.js, while the backend is powered by FastAPI with asynchronous concurrency controls. The core translation engine employs the facebook/nllb- 200-distilled-600M model via HuggingFace Transformers and PyTorch. Speech-to-text is handled by OpenAI Whisper and speech synthesis by Microsoft Edge TTS (edgetts). PDF pro- cessing relies on PyMuPDF (fitz) for spatial geometry analysis, Indic font re-injection, and layout preservation. A multilingual virtual keyboard enables accurate native-script input. Named- entity transliteration via the indic-transliteration library pre- serves proper nouns phonetically, and a gibberish-detection preprocessing layer prevents misleading outputs. Performance is evaluated using BLEU scores, Word Error Rate (WER), and Mean Opinion Score (MOS). The architecture emphasizes modularity, low-resource optimization, and offline adaptability, with ongoing work targeting real-time voice-to-voice translation.},
        keywords = {Multilingual Translation, Neural Machine Translation, NLLB-200, FastAPI, Next.js, Text-to-Speech, Whis- per ASR, PDF Translation, Indian Languages, Indic Transliteration.},
        month = {May},
        }