Saved in:
Bibliographic Details
Main Authors: Woo, Sang Hoon, Lee, Sehun, Kim, Kang-wook, Kim, Gunhee
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.16028
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909797883838464
author Woo, Sang Hoon
Lee, Sehun
Kim, Kang-wook
Kim, Gunhee
author_facet Woo, Sang Hoon
Lee, Sehun
Kim, Kang-wook
Kim, Gunhee
contents Spoken dialogue systems increasingly employ large language models (LLMs) to leverage their advanced reasoning capabilities. However, direct application of LLMs in spoken communication often yield suboptimal results due to mismatches between optimal textual and verbal delivery. While existing approaches adapt LLMs to produce speech-friendly outputs, their impact on reasoning performance remains underexplored. In this work, we propose Think-Verbalize-Speak, a framework that decouples reasoning from spoken delivery to preserve the full reasoning capacity of LLMs. Central to our method is verbalizing, an intermediate step that translates thoughts into natural, speech-ready text. We also introduce ReVerT, a latency-efficient verbalizer based on incremental and asynchronous summarization. Experiments across multiple benchmarks show that our method enhances speech naturalness and conciseness with minimal impact on reasoning. The project page with the dataset and the source code is available at https://yhytoto12.github.io/TVS-ReVerT
format Preprint
id arxiv_https___arxiv_org_abs_2509_16028
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Think, Verbalize, then Speak: Bridging Complex Thoughts and Comprehensible Speech
Woo, Sang Hoon
Lee, Sehun
Kim, Kang-wook
Kim, Gunhee
Computation and Language
Artificial Intelligence
Spoken dialogue systems increasingly employ large language models (LLMs) to leverage their advanced reasoning capabilities. However, direct application of LLMs in spoken communication often yield suboptimal results due to mismatches between optimal textual and verbal delivery. While existing approaches adapt LLMs to produce speech-friendly outputs, their impact on reasoning performance remains underexplored. In this work, we propose Think-Verbalize-Speak, a framework that decouples reasoning from spoken delivery to preserve the full reasoning capacity of LLMs. Central to our method is verbalizing, an intermediate step that translates thoughts into natural, speech-ready text. We also introduce ReVerT, a latency-efficient verbalizer based on incremental and asynchronous summarization. Experiments across multiple benchmarks show that our method enhances speech naturalness and conciseness with minimal impact on reasoning. The project page with the dataset and the source code is available at https://yhytoto12.github.io/TVS-ReVerT
title Think, Verbalize, then Speak: Bridging Complex Thoughts and Comprehensible Speech
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2509.16028