语境相关的音素级语音合成系统中拼接平滑算法

Smoothing algorithm for contextual phone concatenation in speech synthesis

为提高语音合成系统的性能,产生自然流畅的合成语音,该文结合多种拼接点过渡平滑算法,提出了一种以语境相关的音素为基本单元的基于隐Markov(hidden Markov model,HMM)模型的英语拼接合成系统。该合成方法兼有拼接合成以及参数合成的优点,具有相对的灵活性,以及一定的语音自然度。以音素为基本单元尽可能减少了拼接点的个数,降低拼接失真。实验结果表明,多种平滑算法的采用,保证了拼接边界过渡平滑连贯,提高了最终的拼接效果。

A hidden Markov model(HMM)-based English speech synthesis system was built using contextual phone concatenation to improve speech synthesis quality for natural speech.The smoothing algorithm is the key module to ensure both spectrum and prosody continuity between segments.This method flexibly concatenates the segments to generate relatively natural speech.The basic synthesis unit is the phone,which reduces the number of concatenative points and the distortion.A smoothing algorithm is used to smooth the phon...