On the use of phase information for speech recognition

This study addresses the use of short-time phase spectra in automatic speech recognition (ASR). Two recent studies [1,2] have proposed two group delay based spectral representations. Here we propose three new group delay based representations and compare usefulness of all these representations in an ASR experiment. We show that two of the representations we propose perform better, contain equivalent or complementary information to that of the power spectrum and are potentially useful for improving ASR performance.