An improved pair-wise variability index for comparing the timing characteristics of speech

The pair-wise variability index has become a useful and widely used tool for comparing syllable timing of speech. In this paper we present an improved pair-wise variability index based on median instead of means that can more strongly amplify and reveal the differences in the timing characteristics of two datasets. Further, it places less stringent requirements on the pre-processing of the measurements, and it is therefore more robust to outliers in the dataset. The effectiveness of the new measure is demonstrated through an example where the measure is applied to data based on American English speech and Taiwan English speech. The results obtained with the improved pair-wise variability index are compared to those of the standard pair-wise variability index and the rhythm ratio. 1. The pair-wise variability index applied The pair-wise variability index, originally proposed by Low and Grabe [1] and later refined in [2], is a simple measure that allows the rhythm characteristics of different speech samples to be easily compared. In their original study, the authors compared the timing of British English and Singaporean English and showed that there is more variability in consecutive syllable timings for reduced vowels than full vowels in British English than in Singaporean English, which is more staccato sounding. British English reduced vowels therefore have a higher pair-wise variability index than those for Singaporean English, which has a smaller pair-wise variability index. Similar patterns occur for other varieties of English and also in different languages, and the pair-wise variability index has recently been applied by a large number of studies in different domains within the broad field of phonetics. These studies include Whitworth’s study of rhythm in German English [3], Jian’s study of rhythm in Taiwan English [4] and Keane’s studies into rhythm in Tamil [5], [6]. Further, Sandgren [7] used the pair-wise variability index to assess the rhythmic deviations among Swedish learners of French. Variations on the theme include Patel’s study [8] where the pair-wise variability index is used to draw the parallels between the rhythm in language and music. Patel [9] also argues that the term pair-wise variability index is an inaccurate term and proposes to use the term pair-wise contrast index. We choose to use the term pair-wise variability index as this is the most widely known term in the prosodic research community. Other applications of the pair-wise variability index include Williams and Russo’s [10] study into the timing of articulation. Further, Dellwo and Wagner [11] used the pairwise variability index in their study of the relationship between speech rate and rhythm. And Janse [12] used the pair-wise variability index in a study of perception of fast speech. Peter and Stoel-Gammon [13] have applied the pair-wise variability index to speech therapy research, and Kent and Kim [14] have applied it into clinical phonetics research. Other measures of rhythmic variability have also been suggested in the literature, such as the rhythm ratio [15] proposed by Gibbon and Gut. The rhythm ratio is computed as the mean of consecutive syllable duration ratios. Further, Gut [16], [17] applied both the pair-wise variability index and the rhythm ratio to compare a wide range of languages including British English, Nigerian English, Anyi, Ega, Ibibio and German learners of English. The rhythm ratio has currently not been widely embraced in the research community. 2. Improving the pair-wise variability index In acoustical phonetics speech is usually segmented and labeled such that boundaries are marked at instances of time b1, b2, .., bn where n is the number of syllable boundaries. The duration di is therefore computed as follows: i i i b b d − = +1 (1) Based on a sequence of durations the mean pair-wise variability index vii is defined as: