Robust Estimation of Voice Quality Parameters Under Realworld Disturbances

This paper investigates the influence of different types of disturbances to the estimation of voice quality parameters. Here, voice quality is not only based on the pitch or pitch contour as in many approaches. The parameters are estimated by spectral gradients of the vocal tract compensated speech signal. We present a set of five parameters for describing the voice quality. They are used to distinguish between gender, voice qualities, and many emotional states of the speaker. We estimate them from speech signals which are corrupted by background noise and room reverberation. The paper demonstrates a certain degree of robustness of the voice quality parameters against these real world disturbances