English

Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization

Sound 2026-03-16 v1 Computation and Language

Abstract

SpeechLLMs typically combine ASR-trained encoders with text-based LLM backbones, leading them to inherit written-style output patterns unsuitable for text-to-speech synthesis. This mismatch is particularly pronounced in Japanese, where spoken and written registers differ substantially in politeness markers, sentence-final particles, and syntactic complexity. We propose a preference-based alignment approach to adapt Japanese SpeechLLMs for speech-worthy outputs: text that is concise, conversational, and readily synthesized as natural speech. To rigorously evaluate this task, we introduce SpokenElyza, a benchmark for Japanese speech-worthiness derived from ELYZA-tasks-100 with auditory verification by native experts. Experiments show that our approach achieves substantial improvement on SpokenElyza while largely preserving performance on the original written-style evaluation. We will release SpokenElyza to support future research on Japanese spoken dialog systems.

Keywords

Cite

@article{arxiv.2603.12565,
  title  = {Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization},
  author = {Mengjie Zhao and Lianbo Liu and Yusuke Fujita and Hao Shi and Yuan Gao and Roman Koshkin and Yui Sudo},
  journal= {arXiv preprint arXiv:2603.12565},
  year   = {2026}
}
R2 v1 2026-07-01T11:17:46.658Z