What Accounts for Foreign Accent Reduction in Singing?

Abstract

ABSTRACT Second-language learners often sound less foreign-accented when they sing than speak. This study explored whether production rate, particularly vowel lengthening, represents the key factor for accent reduction in song. Stimuli were lyrical lines extracted across spoken and sung versions of popular songs produced by the most-accented male and female non-native speakers from the NUS-48E corpus (Duan et al., 2013). Lengthened versions were also included, either throughout the signal, or for vowels alone, by slowing productions to .67 of the original rate while maintaining fundamental frequency (F0). A total of 33 male and 37 female native English speakers were randomly assigned to rate the level of accentedness for excerpts produced by either male or female speakers. Sung excerpts were perceived as less accented than speech. Unexpectedly, both sung and spoken original excerpts were less accented than both lengthened versions, which provided similar ratings. This suggests that vowel lengthening cannot consistently be responsible for accent reduction in songs. Acoustic analyses revealed a higher median F0 in songs. Median F0 from read and sung excerpts negatively correlated with accentedness ratings, regardless of the speaker's self-identified gender. This suggests that the spreading of adjacent harmonics as F0 increases may under-represent the spectrum, making it difficult for listeners to assess whether vocal tract resonances align with those typically generated by native-language producers.