A review on speech emotion features

Abstract

Research works on combining emotions in intelligent machines are expanding and improving. Human’s speeches basically have various emotional states. The finding of reliable speech features is an ongoing research. Specific features in the speech signal that contribute to emotional information are uncertain, extremely challenging problem and continue being explored. The recognition rate of emotion in speech signal is inconsistent depending on the features used in the experiment and also the database itself. Prosodic, spectral and wavelet features are mostly being used to determine which of these features or its hybrid carry more information about emotions. This paper intends to summarize previous work and make reviews about single and hybrid features based on prosodic, spectral and wavelet feature.