Analysis and Evaluation of Synthetic Data Generation in Speech Dysfluency Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Jinming, Zhou, Xuanru, Lian, Jiachen, Li, Shuhe, Li, William, Ezzes, Zoe, Bogley, Rian, Wauters, Lisa, Miller, Zachary, Vonk, Jet, Morin, Brittany, Gorno-Tempini, Maria, Anumanchipalli, Gopala
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908415583846400
author Zhang, Jinming
Zhou, Xuanru
Lian, Jiachen
Li, Shuhe
Li, William
Ezzes, Zoe
Bogley, Rian
Wauters, Lisa
Miller, Zachary
Vonk, Jet
Morin, Brittany
Gorno-Tempini, Maria
Anumanchipalli, Gopala
author_facet Zhang, Jinming
Zhou, Xuanru
Lian, Jiachen
Li, Shuhe
Li, William
Ezzes, Zoe
Bogley, Rian
Wauters, Lisa
Miller, Zachary
Vonk, Jet
Morin, Brittany
Gorno-Tempini, Maria
Anumanchipalli, Gopala
contents Speech dysfluency detection is crucial for clinical diagnosis and language assessment, but existing methods are limited by the scarcity of high-quality annotated data. Although recent advances in TTS model have enabled synthetic dysfluency generation, existing synthetic datasets suffer from unnatural prosody and limited contextual diversity. To address these limitations, we propose LLM-Dys -- the most comprehensive dysfluent speech corpus with LLM-enhanced dysfluency simulation. This dataset captures 11 dysfluency categories spanning both word and phoneme levels. Building upon this resource, we improve an end-to-end dysfluency detection framework. Experimental validation demonstrates state-of-the-art performance. All data, models, and code are open-sourced at https://github.com/Berkeley-Speech-Group/LLM-Dys.
format Preprint
id arxiv_https___arxiv_org_abs_2505_22029
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Analysis and Evaluation of Synthetic Data Generation in Speech Dysfluency Detection
Zhang, Jinming
Zhou, Xuanru
Lian, Jiachen
Li, Shuhe
Li, William
Ezzes, Zoe
Bogley, Rian
Wauters, Lisa
Miller, Zachary
Vonk, Jet
Morin, Brittany
Gorno-Tempini, Maria
Anumanchipalli, Gopala
Audio and Speech Processing
Artificial Intelligence
Sound
Speech dysfluency detection is crucial for clinical diagnosis and language assessment, but existing methods are limited by the scarcity of high-quality annotated data. Although recent advances in TTS model have enabled synthetic dysfluency generation, existing synthetic datasets suffer from unnatural prosody and limited contextual diversity. To address these limitations, we propose LLM-Dys -- the most comprehensive dysfluent speech corpus with LLM-enhanced dysfluency simulation. This dataset captures 11 dysfluency categories spanning both word and phoneme levels. Building upon this resource, we improve an end-to-end dysfluency detection framework. Experimental validation demonstrates state-of-the-art performance. All data, models, and code are open-sourced at https://github.com/Berkeley-Speech-Group/LLM-Dys.
title Analysis and Evaluation of Synthetic Data Generation in Speech Dysfluency Detection
topic Audio and Speech Processing
Artificial Intelligence
Sound
url https://arxiv.org/abs/2505.22029