Multilingual Stutter Event Detection for English, German, and Mandarin Speech

Felix Haas; Sebastian P. Bayerl

cs.SDMar 27, 2026

Multilingual Stutter Event Detection for English, German, and Mandarin Speech

Authors: Felix Haas, Sebastian P. Bayerl

Organizations: Rosenheim Technical University of Applied Sciences, Germany

Abstract

This paper presents a multi-label stuttering detection system trained on multi-corpus, multilingual data in English, German, and Mandarin.By leveraging annotated stuttering data from three languages and four corpora, the model captures language-independent characteristics of stuttering, enabling robust detection across linguistic contexts. Experimental results demonstrate that multilingual training achieves performance comparable to and, in some cases, even exceeds that of previous systems. These findings suggest that stuttering exhibits cross-linguistic consistency, which supports the development of language-agnostic detection systems. Our work demonstrates the feasibility and advantages of using multilingual data to improve generalizability and reliability in automated stuttering detection.

Figures & tables

Figure 1:

F_{1}

scores relative to the maximum clip lengths included in the training data.

Figure 2: Distribution of clip lengths in the AS - 70 dataset.

#	Test-Set	Bl	Int	Pro	Snd	Wd
Baseline
1	SEP - 28k - E [ 5 ]	0.32	0.77	0.53	0.53	0.64
2	FluencyBank [ 5 ]	0.36	0.79	0.62	0.64	0.52
3	KSoF [ 5 ]	0.64	0.85	0.60	0.48	0.14
4	AS - 70 3 3 3 inference with a model trained using the “Multi-Lingual” training data from [ 5 ]	0.28	0.61	0.60	0.47	0.56
5	AS - 70 [ 14 ]	0.43	0.84	0.70	0.66	0.78

Table 1: Cross-Dataset and multi-lingual dysfluency detection results (

F_{1}

-score) using E2E multi-label systems. ( Bl = Block, Int = Interjection, Pro = Prolongation, Snd = Sound repetition, Wd = Word repetition). Table section headers indicate the training data used, as described in section 4 , followed by the W2V2 base-weights used. Bold values indicate the highest

F_{1}

score for a dysfluency type in one of the datasets.

Figure 3: Distribution in # of distinct samples in the TRILANG-LL training set.

Explore similar work

Multilingual Detection of Alzheimer's Disease from Speech: A Cross-Linguistic Transfer Learning Approach
Jun 4, 2026Nadine Yasser Abdelhalim, Emmanuel Akinrintoyo, Nicole SalomonsSpeech-Based Dementia Detection Speech Processing
Department of Computing, Imperial College London, United Kingdom
Predicting Upcoming Stuttering Events from Three-Second Audio: Stratified Evaluation Reveals Severity-Selective Precursors, and the Model Deploys Fully On-Device
Apr 30, 2026Nazar KozakSpeech Processing Audio Understanding
Kozak Technologies Inc, Los Angeles, CA 90036, USA
Learning from Real Stuttering: Synthetic Mandarin Stuttering Speech for Data Augmentation in Speech Recognition
Jun 19, 2026Yao Lu, Jiaming Zhou, Zhongyuan Fu +1Synthetic Data Augmentation TTS Synthesis
TMCC, College of Computer Science, Nankai University, Tianjin, China

Jun 4, 2026J/K move · Enter open · S save

Multilingual Detection of Alzheimer's Disease from Speech: A Cross-Linguistic Transfer Learning Approach

Nadine Yasser Abdelhalim, Emmanuel Akinrintoyo, Nicole Salomons

Department of Computing, Imperial College London, United Kingdom

Speech-Based Dementia Detection Speech Processing Alzheimer's Disease Classification Cross-Lingual Transfer

Open paper AskPDF arXiv

The development of multilingual Alzheimer's Disease Dementia (AD) detection models presents significant challenges due to the resource-intensive and time-consuming nature of language-specific model training. We propose a novel solution using cross-language training to detect AD in languages beyond those used for model training. This study investigates multilingual deep learning models for detecting AD across different languages and cognitive impairment levels. Using datasets in English, Chinese, Arabic, and Hindi, we developed transformer-based models for binary AD classification. Our approach achieved F1 scores of 82% across all languages, demonstrating strong cross-linguistic generalization. The rapid inference time (0.5 seconds) supports potential real-time screening applications, while consistent performance across languages indicates feasibility for global deployment.

Abstract

Explore similar work

Multilingual Detection of Alzheimer's Disease from Speech: A Cross-Linguistic Transfer Learning Approach

Predicting Upcoming Stuttering Events from Three-Second Audio: Stratified Evaluation Reveals Severity-Selective Precursors, and the Model Deploys Fully On-Device

Learning from Real Stuttering: Synthetic Mandarin Stuttering Speech for Data Augmentation in Speech Recognition