On the Emotion Understanding of Synthesized Speech

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Ge, Yuan, Zhao, Haishu, Hao, Aokai, Zhang, Junxiang, Li, Bei, Liu, Xiaoqian, Wang, Chenglong, Wang, Jianjin, Zhou, Bingsen, Liu, Bingyu, Zhu, Jingbo, Yu, Zhengtao, Xiao, Tong
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915869870784512
author Ge, Yuan
Zhao, Haishu
Hao, Aokai
Zhang, Junxiang
Li, Bei
Liu, Xiaoqian
Wang, Chenglong
Wang, Jianjin
Zhou, Bingsen
Liu, Bingyu
Zhu, Jingbo
Yu, Zhengtao
Xiao, Tong
author_facet Ge, Yuan
Zhao, Haishu
Hao, Aokai
Zhang, Junxiang
Li, Bei
Liu, Xiaoqian
Wang, Chenglong
Wang, Jianjin
Zhou, Bingsen
Liu, Bingyu
Zhu, Jingbo
Yu, Zhengtao
Xiao, Tong
contents Emotion is a core paralinguistic feature in voice interaction. It is widely believed that emotion understanding models learn fundamental representations that transfer to synthesized speech, making emotion understanding results a plausible reward or evaluation metric for assessing emotional expressiveness in speech synthesis. In this work, we critically examine this assumption by systematically evaluating Speech Emotion Recognition (SER) on synthesized speech across datasets, discriminative and generative SER models, and diverse synthesis models. We find that current SER models can not generalize to synthesized speech, largely because speech token prediction during synthesis induces a representation mismatch between synthesized and human speech. Moreover, generative Speech Language Models (SLMs) tend to infer emotion from textual semantics while ignoring paralinguistic cues. Overall, our findings suggest that existing SER models often exploit non-robust shortcuts rather than capturing fundamental features, and paralinguistic understanding in SLMs remains challenging.
format Preprint
id arxiv_https___arxiv_org_abs_2603_16483
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle On the Emotion Understanding of Synthesized Speech
Ge, Yuan
Zhao, Haishu
Hao, Aokai
Zhang, Junxiang
Li, Bei
Liu, Xiaoqian
Wang, Chenglong
Wang, Jianjin
Zhou, Bingsen
Liu, Bingyu
Zhu, Jingbo
Yu, Zhengtao
Xiao, Tong
Computation and Language
Emotion is a core paralinguistic feature in voice interaction. It is widely believed that emotion understanding models learn fundamental representations that transfer to synthesized speech, making emotion understanding results a plausible reward or evaluation metric for assessing emotional expressiveness in speech synthesis. In this work, we critically examine this assumption by systematically evaluating Speech Emotion Recognition (SER) on synthesized speech across datasets, discriminative and generative SER models, and diverse synthesis models. We find that current SER models can not generalize to synthesized speech, largely because speech token prediction during synthesis induces a representation mismatch between synthesized and human speech. Moreover, generative Speech Language Models (SLMs) tend to infer emotion from textual semantics while ignoring paralinguistic cues. Overall, our findings suggest that existing SER models often exploit non-robust shortcuts rather than capturing fundamental features, and paralinguistic understanding in SLMs remains challenging.
title On the Emotion Understanding of Synthesized Speech
topic Computation and Language
url https://arxiv.org/abs/2603.16483