ISCA Archive AVIOS 2012
ISCA Archive AVIOS 2012

Detection of lexical stress using an iterative feature normalization method

Erez Oxman, Eduard Golshtein

Lexical stress plays an important role in the understandability of non-native speakers. It is known that the perception of stress by native English speakers depends on the prosodic features pitch, energy and duration. However, these features are highly variable, and their realization depends on the mother tongue of the speaker. In this paper we present a system for the automatic detection of lexical stress in English disyllabic words spoken by Hebrew and Native English speakers. A novel normalization technique that reduces the variability of the prosodic features is used. This normalization reduces the variability of the intrinsic phrase features such as speaking rate and the intrinsic phoneme features such as phoneme types within the phrase. The pitch gradient feature is used and found to significantly improve performance on Hebrew speakers while having a modest improvement on Native speakers. Detection rates of the system and of non-expert native listeners were compared in reference to that of native expert transcribers. The system achieved primary stress detection rate that outperforms the average detection rate of non-expert listeners.


Cite as: Oxman, E., Golshtein, E. (2012) Detection of lexical stress using an iterative feature normalization method. Proc. Afeka-AVIOS Speech Processing Conference, 33-36

@inproceedings{oxman12_avios,
  author={Erez Oxman and Eduard Golshtein},
  title={{Detection of lexical stress using an iterative feature normalization method}},
  year=2012,
  booktitle={Proc. Afeka-AVIOS Speech Processing Conference},
  pages={33--36}
}