Impact of stimulus difficulty on forced alignment performance in L2 English learners: A focus on L1 Mandarin

This study investigates the performance of forced alignment algorithms and tools for second language (L2) English speech, more specifically L2 English speech produced by L1 Mandarin speakers. A recent study has shown that while the aligner's performance is less consistent with L2 speakers than with native speakers, current forced alignment methods are still accurate enough to be useful for studies of L2 English speech. However, no previous study has investigated whether the difficulty of the spoken content influences accuracy. In the present study, sentences designed to present different levels of difficulty to L1 Mandarin speakers were recorded. Participants included four L1 Mandarin speakers and one L1 English speaker as a control. All participants were associated with a subjective accentedness rating scale by two experienced listeners. A dataset containing recordings was created, which underwent preprocessing, including noise filtering and segmentation by silent periods. Statistical analysis revealed that alignment performance is correlated with accentedness but not with sentence difficulty. The study identified specific phonemic errors and possible differences in variance at different difficulty levels, highlighting the challenges faced by non-native speakers. Forced alignment methods appear to be well suited for phonetic analysis included with L1 Mandarin and L2 English contexts.