HASS: Hierarchical Simulation of Logopenic Aphasic Speech for Scalable PPA Detection

Harrison Li; Kevin Wang; Cheol Jun Cho; Jiachen Lian; Rabab Rangwala; Chenxu Guo; Emma Yang; Lynn Kurteff

eess.ASMar 25, 2026

HASS: Hierarchical Simulation of Logopenic Aphasic Speech for Scalable PPA Detection

Authors: Harrison Li, Kevin Wang, Cheol Jun Cho, Jiachen Lian, Rabab Rangwala, Chenxu Guo, Emma Yang, Lynn Kurteff,

Organizations: UC Berkeley, USA · UCSF, USA · Zhejiang University, China · Columbia University, USA ·

Abstract

Building a diagnosis model for primary progressive aphasia (PPA) has been challenging due to the data scarcity. Collecting clinical data at scale is limited by the high vulnerability of clinical population and the high cost of expert labeling. To circumvent this, previous studies simulate dysfluent speech to generate training data. However, those approaches are not comprehensive enough to simulate PPA as holistic, multi-level phenotypes, instead relying on isolated dysfluencies. To address this, we propose a novel, clinically grounded simulation framework, Hierarchical Aphasic Speech Simulation (HASS). HASS aims to simulate behaviors of logopenic variant of PPA (lvPPA) with varying degrees of severity. To this end, semantic, phonological, and temporal deficits of lvPPA are systematically identified by clinical experts, and simulated. We demonstrate that our framework enables more accurate and generalizable detection models.

Figures & tables

Figure 1: Overview of the HASS hierarchical simulation pipeline.

Severity	Output
Control	Text: The house would go completely dark, save for the single amber glow of a hearth fire. IPA: \textipa D@ h”aUs wUd g”oU k@mpl”iùtli d”Aùôk, s”eIv f3ôD@ s”INg@l ”æmb3ô gl”oU @v@ h”Aùôh f”aI@ô
Mild	Text: The house would go comple \textcolor red [DEL] ly dark, \textcolor blueexcept for the single, \textcolor blueyou know, the ori \textcolor red [SUB] mge light of the, \textcolor bluethe place where you burn the wood, the hearth fire. IPA: \textipa D@ h”aUs wUd gòU k@mpl”iù \textcolor red [DEL] \textipa li d”Aùôk, Eks”Ept f3ôD@ s”INg@l, juù n”oU, DI ”Oôim \textcolor red [SUB] \textipa dZ l”aIt VvD@, D@ pl”eIs wÈô juù b”3ùn D@ w”Ud, D@ h”Aùôh f”aI@ô
Moderate	Text: The house would go, \textcolor blueit would go comple \textcolor red [DEL] ly dark, \textcolor blueexcept for the, the oran \textcolor red [SUB] z light, the am \textcolor red [DEL] ber glow \textcolor bluefrom the, the place where you burn the wood, \textcolor red [PAU] the hea \textcolor red [DEL] th. IPA: \textipa D@ h”aUs wUd g”oU, It wUd gòU k@mpl”iù \textcolor red [DEL] \textipa li d”Aùôk, Eks”Ept f3ôD@, DI ”OôIndz \textcolor red [SUB] \textipa l”aIt, DI ”æm \textcolor red [DEL] \textipa b3ô gl”oU fô VmD@, D@ pl”eIs wÈô juù b”3ùn D@ w”Ud, \textcolor red [PAU] \textipa D@ h”Aù \textcolor red [DEL] \textipa T
Severe	Text: \textcolor blueIt was, it went, uh, \textcolor red [PAU] \textcolor blueno ligh \textcolor red [DEL] , \textcolor bluejust the, the one, the fi \textcolor red [PRO] re thin \textcolor red [SUB] , \textcolor bluein…inside \textcolor red [REP] IPA: \textipa It w”Vz, It w”Ent, ”V, \textcolor red [PAU] \textipa n”oU l”aI \textcolor red [DEL] \textipa , dZ”Vst D@, D@ w”Vn, D@ f”aI \textcolor red [PRO] \textipa @ô T”In \textcolor red [SUB] \textipa , I…Ins”aId \textcolor red [REP]

Table 1: Example output across severity levels for a single ground-truth sentence. Text : word-level dysfluent output. IPA : phoneme-level output with inline markers. Ground truth: “The house would go completely dark, save for the single amber glow of a hearth fire.”

Figure 2: Distribution of phonological dysfluency markers across severity levels. Dashed lines indicate per-severity means.

Model	AUC	F1	Recall (Dys)
Baseline	0.850 $\pm$ 0.122	0.778 $\pm$ 0.165	0.659 $\pm$ 0.238
HASS	0.892 $\pm$ 0.076	0.800 $\pm$ 0.072	0.899 $\pm$ 0.066

Table 2: Cross-site performance comparison (mean

\pm

std).

Figure 3: Comparison of ROC curves for LoRA models trained on real vs HASS speech, evaluated using 5-fold cross-validation. Mean ROC curve and ±1 standard deviation shading.

Figure 4: Cross-site ROC curves for LoRA SFT on w2v2-base. Test data as follows - Group 1: JHU + Capilouto. Group 2: Baycrest + Delaware.

Speech-Based Dementia Detection

Speech-based dementia detection uses spoken language and vocal patterns to identify cognitive impairment and support accessible screening.

See all 44 papers

AphaVoice: Text-to-Speech Model for Aphasia Patient Simulation
Oct 10, 2026Al Varrel Putra Kusuma, Fernando Bello, Joshua BrownTTS Synthesis Clinical Speech Processing
Imperial College London, United Kingdom
Lesioned Multimodal Language Models Reproduce Aphasic Picture-Naming Patterns
Jul 13, 2026Yong Yang, Xiang Guan, Sophie Arheix-Parras +7Multimodal Large Language Models Language Modeling
Department of Computer Science and Engineering, University of South Carolina · Department of Psychology, University of South Carolina · Department of Communication Sciences and Disorders, University of South Carolina +2
LLM-Anchored Paralinguistic Enrichment for Alzheimer's Disease Detection
Sep 9, 2026Xiao Wei, Yuqin Lin, Yaru Cao +6Speech-Based Dementia Detection Speech Prosody
Tianjin Key Laboratory of Cognitive Computing and Application, College of Intelligence and Computing, Tianjin University, Tianjin, China · Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China · College of Computer and Data Science, Fuzhou University, Fuzhou, China +1

Oct 10, 2026J/K move · Enter open · S save

AphaVoice: Text-to-Speech Model for Aphasia Patient Simulation

Al Varrel Putra Kusuma, Fernando Bello, Joshua Brown

Imperial College London, United Kingdom

TTS Synthesis Clinical Speech Processing

Open paper AskPDF arXiv

While language therapy is the primary treatment for post-stroke aphasia, limited clinical resources often restrict patient access to proper care. Addressing this gap requires reliable training simulations for medical professionals. However, current AI efforts focus almost exclusively on speech processing, leaving speech generation largely unexplored. This study introduces AphaVoice, a novel text-to-speech (TTS) framework explicitly engineered to synthesize highly realistic aphasic speech. Evaluated against state-of-the-art TTS systems, AphaVoice demonstrates a superior ability to capture the clinical feature of dysfluency, outperforming existing models in temporal metrics while consistently maintaining speaker identity. By leveraging latent space shifting, the model also supports multi-speaker generation without sacrificing phonetic realism. Ultimately, AphaVoice provides a scalable, high-fidelity tool for medical training and clinical simulation related to aphasia.

Abstract

Speech-Based Dementia Detection

Explore similar work

AphaVoice: Text-to-Speech Model for Aphasia Patient Simulation

Lesioned Multimodal Language Models Reproduce Aphasic Picture-Naming Patterns

LLM-Anchored Paralinguistic Enrichment for Alzheimer's Disease Detection