汉语语音韵律强度的量化计算方法

Quantitative Calculation of Prosodic Strength in Mandarin Speech

韵律强度是指连续语流中各音节相对突显度的大小。以往对韵律强度的研究常常根据听感将其标注为几个强弱不同的等级,然而,基于听感的标注过于主观,且无法给出连续变化的数值,不能很好地满足语音技术的需求。为此,本文提出了一个计算汉语连续语流每个音节韵律强度的方法,基于三个声学度量(归一化的音节时长、基频扩张度、调形与模板的偏离度)做线性估算。通过分析基于声学参数算出的韵律强度值与基于听感标注的重音指数之间的关系、韵律强度值与词性之间的关系以及音节在焦点与非焦点条件下韵律强度值的差异,初步验证了该模型的有效性。

Prosodic strength refers to the relative prominence of each syllable in continuous speech. It used to be perceptually annotated at discrete degrees,but perceptual annotation is highly subjective and cannot give continuous values that are potentially useful in speech technologies. This study proposed a method to estimate prosodic strength of each syllable in Mandarin continuous speech,using a linear combination of three acoustic measures,viz.,normalized syllable duration,pitch span,and F0 deviation from tone template. The validity of the method was verified by the relationship between the acoustically estimated prosodic strength and the perceptually labeled stress index,by the relationship between prosodic strength and POS class,and by the comparison on prosodic strengths of the same words in different focal conditions.