Comparison of time and frequency domain measures of the voice source

Abstract

Voice source modulation is of fundamental importance in speech communication. Many parameters have been proposed to capture the characteristics of the voice, but it is not always clear how the different kinds of parameters relate to each other. Two widely used approaches to voice parameterisation involve either time or frequency domain measures. In this paper they are compared in the analysis of a small data set – an utterance produced with different affects. The results for the time and frequency parameterisation are correlated, and both are compared in turn to an alternative parameter, MDQ, which is based on the wavelet transform. The correlation of time and frequency measures was reasonably high for parameters that relate to the lower end of the source spectrum, but correlations for the parameters relating to the upper end of the spectrum ran contrary to expectations. MDQ measures correlated strongly with time domain measures, and reasonably strongly with frequency domain measures of the low end of the source spectrum.