计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (1): 168-171.DOI: 10.3778/j.issn.1002-8331.1503-0314

• 模式识别与人工智能 • 上一篇    下一篇

后置滤波器参数自适应的语音合成改进算法

戈永侃,于凤芹   

  1. 江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2017-01-01 发布日期:2017-01-10

Improved speech synthesis with adaptive postfilter parameters

GE Yongkan, YU Fengqin   

  1. School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2017-01-01 Published:2017-01-10

摘要: 针对HMM语音合成算法,固定参数的后置滤波器无法适应不同失真程度的频谱导致合成语音自然度下降,提出了一种基于后置滤波器参数自适应的语音合成改进算法。该方法根据语音谱的平坦度自适应选择最优的短时滤波参数来对合成语音频谱的共振峰区域增强;使用长时后置滤波器优化合成语音的基频谐波结构来减轻合成语音基频的不连续性。仿真实验结果表明,该方法能够有效地减轻语音的频谱过平滑,主观测试结果表明,合成语音的自然度得以提高。

关键词: 语音合成, 后置滤波器, 参数自适应, 自然度, 隐马尔可夫模型(HMM)

Abstract: In order to overcome the decrease of speech naturalness resulting from fixed postfilter parameters, the improved speech synthesis with adaptive postfilter parameters is proposed. Short-term postfilter parameters are adapted to variations in spectral flatness that obtained from the speech to enhance the formants of spectrum and a long-term postfilter is used to emphasize the signal components at harmonic peaks and attenuate the discontinuity of pitch. Simulation experiment results demonstrate that the method can alleviate spectral over-smoothing and subjective tests show that the speech naturalness is improved.

Key words: speech synthesis, postfilter, adaptive parameter, naturalness, Hidden Markov Model(HMM)