Singing Your Accent Away, and Why It Works

ABSTRACT Eleven Dutch secondary school children sang and spoke the lyrics of a number of popular English songs. Their pronunciation accuracy was rated significantly higher in the singing mode by a group of native speakers. Additional perception scores showed that the decrease in Dutch-accentedness must be due to the fact that their foreign accent could not be signaled in the f0 and segment durations during singing. Keywords: foreign accent, singing, Dutch accented English 1. INTRODUCTION Adults learning a foreign or second language are typically perceived as having an accent by native speakers of the language. A great number of studies have investigated the concept of foreign accent (e.g., [4, 5, 7, 10, 13]), many of which have dealt with the perception of foreign-accentedness, and the factors that affect it, like speaking rate [11, 12]. To the best of our knowledge, the effect of singing on the perception of foreign-accentedness has never been investigated. It is this factor which the present study addresses. Sounds (and other units of speech) may differ from one another in quality, pitch, loudness and duration [9]. Non-native speakers are often recognized by native speakers because of deviations from native norms for those phonetic dimensions. Our concern is not to explain foreign-accentedness (e.g., [1, 2, 6, 8]), but to show that deviations from native norms during singing only concern quality and loudness and that as a result foreign-accentedness is reduced when non-native speakers sing. A number of differences between speech and singing have been observed. Professional Western opera singers modify the quality of their vowels to avoid masking by the loud orchestral accompaniment they generally have to sing to. Male professional Western opera singers accomplish this by bringing F3, F4, and F5 closer together, increasing loudness and audibility. This is also characteristic of male Western concert singing [15, 16]. The modifications made by female Western opera singers are different from those described above; when F1 is lower than f0, F1 is brought closer to f0, increasing audibility [16]. More generally, the pitch contour in singing is determined by the composer [16] and the same is true for duration. Sundberg [17] points out another difference between singing and speech when he notes that pitch changes in singing are perceived differently from those in speech. While pitch change size in speech is perceived continuously, pitch changes in singing are categorized into musical intervals. Also, tone length differences in singing are perceived categorically, i.e., as note values [17]. But the most important and general difference between speech and singing for our purposes is that singing imposes duration and f0 patterns on speech. Consequently, the usual segmental durations and f0 variation of speech are not available in singing. Anecdotally, people have reported that when they sing, they seem to have a less strong foreign accent than when they speak. This contribution investigates the effect of singing on the perception of Dutch-accentedness in English. We hypothesize, first, that Dutch speakers of English have a less strong foreign accent when they sing than when they speak. At first blush, a number of explanations might be put forward if this were found to be true. One is that the realization of segments is more authentic in singing than in speech. Another might be that listeners judge foreign-accentedness in singing less harshly than that in speech. A third candidate, one which we consider the most likely explanation, is that in singing, the syllable durations and f0 patterns of spoken texts are imposed by the rhythm and melody of the musical score. As a result, the conventional durations and f0 of spoken texts are no longer available, and thus neither is their contribution to the degree of foreign-accentedness.