Rhythm metrics of spontaneous speech and accent

Abstract

The present study reports preliminary results of an experimental evaluation of the stability of rhythm metric scores across different speech genres (i.e., read sentences vs. spontaneous speech) in English and Japanese. Data for native English (L1 English) speakers and from one L2 English speaker (L1 Japanese) was extracted from the Wildcat Corpus (Van Engen et al., 2010). L2 English and L1 Japanese data was extracted from unpublished data generated in an active project investigating spontaneous speech across dialects and accents (Warner et al., 2015). Three metrics (i.e., %V, VarcoV, and nPVI_V) were employed to quantify durational characteristics of the total 48 spontaneous utterances (L1 English: 3 speakers x 3 utterances, L1 Japanese: 3 speakers x 3 utterances, L2 English: 6 speakers x 5 utterances). The rhythm metric scores from the spontaneous corpora were compared to the results in Grenon and White (2008) who examined English and Japanese speech rhythm with 90 read sentences. We predict that measures of rhythm will differentiate between read and spontaneous speech, with spontaneous speech falling on the faster side of each metric. While Grennon and White (2008) had difficulty distinguishing between L1 Japanese and L2 English speech, we predict that using more natural speech will allow for better discrimination of these two speech groups. As predicted, the spontaneous speech had a slightly lower %V scores and a slightly higher VarcoV scores as compared to the read speech of Grenon and White (2008). However, we found that the speech rate of the spontaneous speech was slower than the read speech. The results suggest that metric scores pattern in a similar way across of speech genres, but that the spontaneous speech better distinguishes the L1 Japanese from the L2 English.