BROWSE
Detail
Conference Paper
Bidirectional Spoken-Written Text Conversion with Large Language Models
Cited 0 time in 
Share 
Share
- Authors
- Muyeol Choi, HyunJung Choi, Yohan Lim, Jeonguk Bang, Minkyu Lee, Seonhui Kim, Seung Yun, Donghyun Kim, Minsoo Kim, SangHun Kim
- Issue Date
- 2025-08
- Citation
- International Speech Communication Association (INTERSPEECH) 2025, pp.5088-5092
- Language
- English
- Type
- Conference Paper
- Abstract
- Traditional ASR systems normalize transcriptions into spoken form for training, leading to spoken form outputs. In contrast, modern Transformer-based models directly map speech-to-text, preserving the written form. However, existing speech databases mainly contain spoken form, causing inconsistencies in recognition results. To address this, Inverse Text Normalization (ITN) is required, and dual transcription data is crucial for effective training. However, constructing such datasets is costly and time-consuming. This study proposes a data augmentation method leveraging LLMs to automatically generate dual transcription data with minimal effort. It employs iterative learning to expand the dataset through supervised and semi-supervised methods. Additionally, the bidirectional text conversion (BTC) model supports both ITN and TN within a unified framework. Experimental results demonstrate that the proposed method achieved an ERR of 13.4% and 4.7%, outperforming prior approaches.
- KSP Keywords
- ASR Systems, Augmentation method, Data Augmentation, Effective training, Semi-supervised, Speech-To-Text(STT), Text normalization, iterative learning, language models, supervised method, transformer-based