A hybrid LLM-augmented Bangla text normalization framework for text-to-speech.
Kingfisher is a three-stage hybrid Bangla text normalization framework for Text-to-Speech, combining LLM-based tokenization and semiotic class annotation, lexicon-driven context-aware verbalization, and automated error correction.
It achieves 96% overall accuracy (95% CI: 95-97%) across diverse Bangla texts, outperforming Sparrowhawk, the only previously available Bangla normalizer. Kingfisher is released together with the first public Bangla text normalization dataset.
Kingfisher is a three-stage hybrid Bangla text normalization framework for Text-to-Speech, combining LLM-based tokenization and semiotic class annotation, lexicon-driven context-aware verbalization, and automated error correction. It achieves 96% overall accuracy (95% CI: 95-97%) across diverse Bangla texts, outperforming Sparrowhawk, the only previously available Bangla normalizer.
@article{kingfisher2026,title={Kingfisher: A hybrid LLM-augmented Bangla text normalization for enhanced text-to-speech},author={Raju, Rajan Saha and Ahmad, Arif and Rahman, Mohammad Shahidur},journal={Natural Language Processing Journal},year={2026},month=feb,}