CASIA OpenIR  > 模式识别国家重点实验室  > 语音交互
基于自适应训练的疑问句语音合成
方硕; 陶建华
2015-10
Conference Name第十三届全国人机语音通讯会议
Source Publication第十三届全国人机语音通讯会议
Conference Date2015-10
Conference Place天津
Abstract针对目前合成语音缺乏表现力的现状,本文提出了一种基于自适应训练的疑问句语音合成方法。采用基于统计参数语音合成技术,用大规模的陈述句语料训练初始声学模型,在此基础上,采用小规模的疑问句语料进行自适应训练,得到疑问句的声学模型,从而合成出具有疑问语气的语音。实验结果表明,在疑问句训练语料较少的情况下,该方法的合成语音在自然度以及客观评价指标上明显优于以相同疑问句语料直接训练的方法,并且能用较少的语料达到直接采用大语料训练的效果
Other AbstractIn order to improve the expressiveness of synthetic speech, a question speech synthesis method based on adaptive training is proposed in this paper. By statistical parametric speech synthesis based on MSD-HSMM, a large corpus of exclamatory sentences is used to train the initial acoustic model. Based on the initial acoustic model, a small corpus of question sentences is adaptively trained to gain the question acoustic model and then synthesize the question speech. Conclusions can drew from the experiments that the subjective and objective evaluations of the synthetic speech by this method is superior to the traditional method, where the same size corpus of question sentences is trained directly. Besides, the results of a relatively small training corpus by this method are comparable with the larger training corpus by the traditional method
Keyword疑问句生成 自适应 语音合成
Indexed By其他
Document Type会议论文
Identifierhttp://ir.ia.ac.cn/handle/173211/11773
Collection模式识别国家重点实验室_语音交互
Corresponding Author陶建华
Affiliation中国科学院自动化研究所
Recommended Citation
GB/T 7714
方硕,陶建华. 基于自适应训练的疑问句语音合成[C],2015.
Files in This Item: Download All
File Name/Size DocType Version Access License
基于自适应训练的疑问句语音合成 _rev(771KB)会议论文 开放获取CC BY-NC-SAView Download
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[方硕]'s Articles
[陶建华]'s Articles
Baidu academic
Similar articles in Baidu academic
[方硕]'s Articles
[陶建华]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[方硕]'s Articles
[陶建华]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: 基于自适应训练的疑问句语音合成 _review.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.