一种混合使用连续HMM和离散的合成系统.pdfVIP

一种混合使用连续HMM 和离散 HMM 的HTS 合成系统 [1] [2] [3] [4] 于剑 、张蒙 、陶建华 、王霞 [1] [2] [3] （中科院自动化所模式识别国家重点实验室，北京100080） [4] （诺基亚中国研究中心，北京） 摘要：传统的基于隐马尔可夫模型（HMM ）的语音合成系统（HTS ）存在两个问题，分别是时域的过平滑问题和频域的 过平滑问题。为了解决这两个问题，本文提出了一种同时利用连续隐马尔可夫模型和离散隐马尔可夫模型的合成系统框架。 传统的HTS 合成系统使用高斯函数的均值表征频谱包络，这不可避免地带来了频域过平滑问题，而本文所提出系统通过 使用来自于真实频谱的离散码本矢量表征频谱包络，有效地克服了频域上的过平滑问题；传统HTS 系统利用动态参数生 成完整频谱，导致生成的频谱在时域上缺少变化，而本文则提出一个精心设计的码本矢量选取算法来生成完整频谱，以此 克服时域过平滑问题。实验证明，通过混合使用连续HMM 和离散HMM ，合成语音的清晰度可以得到极大提高。 关键词：语音合成，隐马尔可夫模型，语音处理 A Novel HTS System Using both Continuous HMMs and Discrete HMMs Jian Yu, Meng Zhang, Jianhua Tao, Xia Wang (National Laboratory of Pattern Recognition, Chinese Academy of Science, Beijing, 100080) (Nokia Research Center, Beijing) Abstract: The conventional HMM-based speech synthesis system (HTS) has encountered two over-smoothing problems in both time domain and frequency domain. To resolve these two problems, this paper presents a new HTS framework using both continuous HMMs and discrete HMMs. By the replacement of spectral envelope represented by continuous Gaussian distribution with that represented by discrete codevector, the over-smoothing problem in frequency domain can be resolved, and by the replacement of the parameter generation algorithm using dynamic features with a new well-designed codevector selection algorithm, the over-smoothing problem in time domain can also be better resolved. Experimental results show that the using of both continuous HMMs and discrete HMMs significantly improves the voice quality of synthesized speech. Keyword: Speech Synthesis, Hidden Markov Model, Speech Processing. 1．引言 成频谱参数在时域上的多样性[4]；（2 ）应用一个被称为