Relationship between acoustic features and speech intelligibility

Clear speech is more intelligible than conversational speech, but the specific acoustic features responsible are not fully understood. This work uses a hybridization method to replace individual conversational speech features with clear speech features to measure their impact on intelligibility. Experiments show that modifying duration and spectral features, as well as formant frequencies, improves intelligibility, while changes to energy, F0, phoneme sequence, and pause patterns do not. A formant contour model was also developed to analyze formant targets and coarticulation, achieving low fitting errors and supporting improved formant‑tracking accuracy.