arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.03541cs.CLcs.LGcs.SDeess.AS
收藏

用于可扩展低资源语音合成的多语言Byte2Speech模型

Multilingual Byte2Speech Models for Scalable Low-resource Speech Synthesis

The Hong Kong University of Science and Technology(香港科技大学);Microsoft China(微软中国)

Mutian He, Jingzhou Yang, Lei He, Frank K. Soong

摘要:

为了将神经语音合成扩展到各种真实世界的语言,我们提出了一个多语言的端到端框架,该框架将字节输入映射到频谱图,从而允许任意输入脚本。除了在40多种语言上的强大结果外,该框架还展示了在极端低资源甚至只有40秒转录记录的Few-Shot场景下适应新语言的能力,而不需要词典,额外语料库,辅助模型或语言专业知识等每种语言资源,从而确保可扩展性。虽然它保留了令人满意的可理解性和自然匹配丰富的资源模型。详尽的比较和消融研究,以揭示低资源语言的框架的潜力。此外,我们提出了一种新的方法来提取特定语言的子网络的多语言模型,以更好地理解其机制。

英文摘要:

To scale neural speech synthesis to various real-world languages, we present a multilingual end-to-end framework that maps byte inputs to spectrograms, thus allowing arbitrary input scripts. Besides strong results on 40+ languages, the framework demonstrates capabilities to adapt to new languages under extreme low-resource and even few-shot scenarios of merely 40s transcribed recording, without the need of per-language resources like lexicon, extra corpus, auxiliary models, or linguistic expertise, thus ensuring scalability. While it retains satisfactory intelligibility and naturalness matching rich-resource models. Exhaustive comparative and ablation studies are performed to reveal the potential of the framework for low-resource languages. Furthermore, we propose a novel method to extract language-specific sub-networks in a multilingual model for a better understanding of its mechanism.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑