UtterTune: LoRA-Based Target-Language Pronunciation Edit and Control in Multilingual Text-to-Speech
arXiv:2508.09767v3 Announce Type: replace-cross Abstract: We propose UtterTune, a lightweight method for adapting a multilingual text-to-speech (TTS) system built on a large language model (LLM). It improves control of pronunciation in the target language while preserving performance in the others. Although LLM architectures have enabled TTS models to achieve remarkable naturalness, accurately modeling grapheme-to-phoneme (G2P) mapping and prosody remains challenging, especially when the model omits an explicit G2P module and directly processes minimally encoded text (e.g., byte-pair encoding)
The rapid advancement of large language models for text-to-speech necessitates solutions for fine-grained control and accuracy, especially in multilingual contexts where pronunciation nuances are critical.
Improving target-language pronunciation control in multilingual TTS systems is crucial for more natural and globally applicable AI voice technologies, impacting user experience and accessibility across diverse linguistic landscapes.
Multilingual TTS models can now achieve more accurate and controlled pronunciation without sacrificing performance in other languages, enhancing their versatility and adoption.
- · Multilingual AI platforms
- · Global content creators
- · Language learning applications
- · LLM developers
- · Monolingual TTS solutions
- · Less precise multilingual AI systems
Enhanced multilingual voice synthesis will lead to more natural-sounding AI assistants and automated services.
Improved pronunciation accuracy will lower linguistic barriers in global communication and content consumption.
The perceived 'foreignness' of AI-generated speech could diminish, fostering greater human-AI interaction in diverse cultural settings.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL