A Comparison of Rhythm Metrics for L2 Speech

A wide range of rhythm metrics (global metrics: %V, Δ, Varco, and segVarco; pairwise metrics: rPVI, nPVI, CCI, and D nCCI) was applied to L1 Japanese speakers’ L2 English speech data. Less proficient Japanese speakers of English are expected to show less durational variability for both vocalic and consonantal intervals (because of insufficient stress realization and transfer of CV syllable structure), although this pattern may be obscured by their slower speech rate (which increases interval durations in general). To test if the metrics can capture the L2 rhythmic characteristics, each metric was applied to read speech samples of “The North Wind and the Sun” by 183 Japanese speakers in the J-AESOP corpus. Only %V, VarcoV, and segVarcoV/C were successful; other metrics yielded inconsistent or implausible results likely due to insufficient rate normalization. The overall results indicate that global metrics can effectively quantify L2 rhythm if speech rate is normalized by the mean duration of segments (which is a good predictor of tempo) rather than the mean interval duration (which is popular but susceptible to syllable complexity).