Using Nonlinear Features in Automatic English Lexical Stress Detection

Abstract

Lexical stress is an important prosodic feature, especially for stress-timed language such as English. This paper proposes three novel features, based on the nonlinear Bark scale and the Teager Energy Operator (TEO), for automatic English lexical stress detection. The proposed features are Bark Scale Cepstrum (BSC), Time Domain TEO-Bark Scale Cepstrum (TDT-BSC) and Frequency Domain TEO-Bark Scale Cepstrum (FDT-BSC). Their contributions, along with traditional features and their combinations, to English lexical stress detection are evaluated by single word pairs and continue sentences. Evaluation results showed that these new features gave significant improvement over traditional ones.