Phonemizing informal text improves pronunciation in speech systems
Phonemizing User-Generated Text: A Benchmark, Taxonomy, and Compositional Approach
Computation and LanguageArtificial Intelligence
Summary
Text-to-speech systems often struggle to pronounce informal user-generated text like 'ppl' or 'imo' correctly because these words are shortened or spelled differently from their usual forms. The authors created a new benchmark called UGTPhon to test how well current models handle this challenge in English, Vietnamese, and Korean. They found that existing models make many mistakes when converting these informal words to their spoken forms. To improve this, they designed a method that uses the standard word forms as hints during pronunciation. This approach reduces errors and can compete with some larger language models while using smaller ones.
What this means in practice
- •For speech technology developers: Improve pronunciation accuracy of user-generated text inputs in text-to-speech systems by incorporating canonical form lookup in grapheme-to-phoneme conversion.
- •For multilingual chatbot builders: Enhance understanding and pronunciation of informal user text in English, Vietnamese, and Korean for more natural vocal responses.
Authors
MinJu Jeon, Younghan Park, Han Sung Park, Jong-Hwan Kim, Dong-Jin Kim, Hoyeon Lee
Abstract
Text-to-speech systems increasingly process user-generated text (UGT) such as ppl and imo, whose pronunciation must be inferred from the canonical rather than surface form. We introduce UGTPhon, the first grapheme-to-phoneme (G2P) benchmark for UGT in English, Vietnamese, and Korean, together with an inference-grounded taxonomy for fine-grained diagnosis. Existing G2P models and frontier LLMs exhibit a systematic canonical-to-non-canonical performance gap, reaching up to 66.8 PER points. As a benchmark baseline, we propose a simple compositional G2P approach that incorporates canonical-form evidence through exact-match lookup and staged decoding. Across matched ByT5 and Qwen2.5-0.5B backbones, explicit canonical-form modeling consistently reduces non-canonical G2P errors. The 0.5B variant also performs competitively with much larger few-shot frontier LLMs, highlighting the benefit of explicitly modeling canonical-form inference for UGT phonemization.