Investigating Holistic Measures of Speech Prosody through Ratings of Filtered Speech

Prosody has been demonstrated to be a key component in second language acquisition and assessment. This study analyzes raters’ ability to distinguish different levels of L1 and L2 prosody when listening to filtered speech. The goal of this study is to investigate how many levels of prosodic proficiency can be made when it is isolated in a filtered speech condition. An experiment was conducted with 45 L2 speech sound clips that had been recorded as a part of the OEPT (Oral English Proficiency Test). The test items used for this project included describing a graph in English and reading aloud an academic passage. These L2 English sound clips were produced by native Mandarin speakers at the novice, intermediate, and advanced levels of English proficiency. For control purposes, another 30 sound clips were recorded of L1 Mandarin and L1 English speakers completing the same task each in their native language. All sound clips were filtered with a low pass filter of a 500 Hz cutoff in Praat 5.1.18. The results indicated a high correlation between the raters’ scores in the filtered condition (listening primarily to prosodic components) and the original scores assigned by raters in an unfiltered condition. This reveals a strong link between the stages of second language prosody acquisition and overall L2 proficiency ratings. The results demonstrate that trained raters’ can distinguish typologically distinct prosodic systems when listening to filtered speech, as well as make reliable judgments concerning the L2 prosodic proficiency level.