基于GPT-2和互信息的语言单位信息量对韵律特征的影响(Prosodic Effects of Speech Unit’s Information Based on GPT-2 and Mutual Information)

Yun Hao (郝韵), Yanlu Xie (解焱陆), Binghuai Lin (林炳怀), Jinsong Zhang (张劲松)


Abstract
“基于信息论的言语产出研究发现携带信息量越大的语言单位,其语音信号越容易被强化。目前的相关研究主要通过自信息的方式衡量语言单位信息量,但该方法难以对长距离的上下文语境进行建模。本研究引入基于预训练语言模型GPT-2和文本-拼音互信息的语言单位信息量衡量方式,考察汉语的单词、韵母和声调信息量对语音产出的韵律特征的影响。研究结果显示汉语中单词和韵母信息量更大时,其韵律特征倾向于被增强,证明了我们提出的方法是有效的。其中信息量效应在音长特征上相比音高和音强特征更显著。”
Anthology ID:
2022.ccl-1.5
Volume:
Proceedings of the 21st Chinese National Conference on Computational Linguistics
Month:
October
Year:
2022
Address:
Nanchang, China
Venue:
CCL
SIG:
Publisher:
Chinese Information Processing Society of China
Note:
Pages:
46–55
Language:
Chinese
URL:
https://aclanthology.org/2022.ccl-1.5
DOI:
Bibkey:
Cite (ACL):
Yun Hao, Yanlu Xie, Binghuai Lin, and Jinsong Zhang. 2022. 基于GPT-2和互信息的语言单位信息量对韵律特征的影响(Prosodic Effects of Speech Unit’s Information Based on GPT-2 and Mutual Information). In Proceedings of the 21st Chinese National Conference on Computational Linguistics, pages 46–55, Nanchang, China. Chinese Information Processing Society of China.
Cite (Informal):
基于GPT-2和互信息的语言单位信息量对韵律特征的影响(Prosodic Effects of Speech Unit’s Information Based on GPT-2 and Mutual Information) (Hao et al., CCL 2022)
Copy Citation:
PDF:
https://preview.aclanthology.org/ingestion-script-update/2022.ccl-1.5.pdf