Mandarin word duration and pitch predicted by contextualized embeddings

Researchers have developed a method using contextualized embeddings (CEs) to predict spoken word duration and pitch in Mandarin monosyllabic words. The study demonstrated that CEs can accurately predict word duration, even at the individual token level, outperforming chance and permutation baselines. These predicted durations were precise enough to reconstruct pitch contours, which also approximated empirical contours and surpassed a permutation baseline. AI

IMPACT This research could improve speech synthesis and recognition systems by enabling more accurate prediction of prosodic features.

RANK_REASON Academic paper detailing a new method for predicting linguistic features using embeddings. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Mandarin

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →