【摘要】 在现有参数语音合成系统中,参数合成器本身的局限影响了语音质量的提高。尝试改进传统的正弦分析合成器,引入STRAIGHT分析合成器中去除周期影响的谱估计思想,在最小均方误差准则下,更准确而高效的估计出正弦参数。同时通过去除线性相位和相位解卷绕,实现相位谱上音高和时长的调整。实验结果表明,正弦合成器对比STRAIGHT,重建语音的质量更高,同时具有灵活的音高时长调整能力。该分析合成器具备了在嵌入式和其他合成系统中应用的潜力。更多还原

【Abstract】 In existing parametric speech synthesis systems,the main difficulty for improving speech quality comes from the limitations of parametric synthesizers.This paper tries to improve traditional sinusoidal analyzer by introducing STRAIGHT’s spectrum estimation idea of removing the effect of signal periodicity.The sinusoidal parameters’ extraction becomes more efficient and precise under MSE criterion.Time-scale and pitch modification on phase spectrum are presented by removing linear phase and phase unwrapping.The experimental result shows that this sinusoidal synthesizer reproduces speech with better quality than STRAIGHT and good capability for prosodic modification.This synthesizer obtains the potential for the implementation in embedded and other synthesis systems.更多还原