Measures of tone difficulty

Abstract

Measures of difficulty are needed in many real life as well as computer-simulated applications. Such measures for text, math and science have long received academic and industrial attention due to the demands for k-12 instruction and assessment. In recent years, the demands for comparable studies of speech are on the rise given the popularity of on-line second language teaching software and games. The goal of this project is to explore whether the acoustic attributes of Mandarin lexical tones obtained from individual sound files can explain their level of difficulty experienced by second language learners. The study uses monosyllabic tones, thus isolating the task from measurements of complexity in order to focus on acoustics. We recorded sound files that are rich in natural variation using different levels of talker-to-listener distance, used quadratic decomposition to obtain three coefficients that represent each tonal contour, and analyzed their relationship with learners' performance. We will report the difference between native and second-language tone perception. The results have potential applications in speech synthesis: to generate tone tokens with different levels of difficulty for language learners, and different levels of talker-to-listener distance.