Research

Paper

AI LLM March 13, 2026

Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization

Authors

Mengjie Zhao, Lianbo Liu, Yusuke Fujita, Hao Shi, Yuan Gao, Roman Koshkin, Yui Sudo

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.

Metadata

arXiv ID: 2603.12565
Provider: ARXIV
Primary Category: cs.SD
Published: 2026-03-13
Fetched: 2026-03-16 06:01

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.12565v1</id>\n    <title>Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization</title>\n    <updated>2026-03-13T01:55:18Z</updated>\n    <link href='https://arxiv.org/abs/2603.12565v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.12565v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>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.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.SD'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-13T01:55:18Z</published>\n    <arxiv:primary_category term='cs.SD'/>\n    <author>\n      <name>Mengjie Zhao</name>\n    </author>\n    <author>\n      <name>Lianbo Liu</name>\n    </author>\n    <author>\n      <name>Yusuke Fujita</name>\n    </author>\n    <author>\n      <name>Hao Shi</name>\n    </author>\n    <author>\n      <name>Yuan Gao</name>\n    </author>\n    <author>\n      <name>Roman Koshkin</name>\n    </author>\n    <author>\n      <name>Yui Sudo</name>\n    </author>\n  </entry>"
}