Page 13 - Summer 2018

Figure 1. D' estimates for native listeners detecting whether different kinds of speech samples were produced by a nonnative speaker. The speech samples were produced by English speak- ers from the United States and nonnative speakers from Korea. Data from Park (2008).
number of measurements to quantify many perceptually rel- evant aspects of speech acoustics, especially those aspects that are known to distinguish words in various languages. Howev- er, the level of informational detail indicating a foreign accent is at some finer-grained level than what is used to characterize phonetic behavior with our current instrumentation.
One takeaway from these results might be that vowels are very important carriers of speech information, within which many very fine shadings of relevant information are available. For example, there is a common observation that vowel informa- tion is heavily responsible for the differences between English dialects, leading to a well-developed tradition of comparing spectral vowel shapes between speakers of different dialects to map out the geographic variation between the various dia- lects. There is very extensive research laying out vowel infor- mation in a broad range of dialect research, including work published in Acoustics Today (Jacewicz and Fox, 2016) and in the enormous volume of material published in the Atlas of North American English (Labov et al., 2006). For more target- ed research of this type, see Clopper et al. (2005).
However, this kind of fine detail is not just available in vow- els. Flege (1984), in a classic article published in The Journal of the Acoustical Society of America, reported a series of ex- periments on the perception of foreign accent in nonnative speakers of English living in Birmingham, Alabama. The na- tive language of the speakers was French. Across the experi- ments, Flege varied both the content of the speech itself and the experience level of the listeners.
Figure 2. Proportion of time a stimulus was identified as being produced by a nonnative speaker in four of the experiments reported in Flege (1984).
Figure 3. A spectrographic representation of a native produc- tion of the phrase “Two Little Boys.” Solid red lines, portion of the spectral image associated with the text at the top; dashed cursors and the three lines of text at the bottom, the three stim- ulus types that yielded the accent perception rates given in Fig- ure 2, the largest being the phrase, the next being the syllable, and the final type being just the first 30 ms of the production.
In the first experiment, Flege (1984) began by having pho- netically trained listeners with experience in French listen- ing to entire phrases. Here, he found very high rates of ac- curacy. Through four experiments, Flege progressively used listeners with less training and presented shorter and shorter bits of speech to these listeners. In the last experiment, he collected phonetically untrained undergraduate students without extensive French experience and presented to them just the burst release from the consonant /t/.
Summer 2018 | Acoustics Today | 11