Statistical evaluation of the effect of gender on prosodic parameters and their influence on gender-dependent speech recognition

An important issue to be taken into account in automatic speech recognition (ASR) is the influence of gender on the recognition process. Obvious differences in the speech uttered by females versus that of males are the main reasons for the differences in their speech features. Therefore, the behavior of the speech recognition systems is, to some extent, related to the speaker's gender. In this paper, the effect of speaker's gender on the prosodic parameters of Persian (Farsi) speech is evaluated. Then, taking into account the speaker's gender, the effect of prosodic parameters, such as pitch and formant frequencies, on recognition results is inspected. For this, prosodic parameters are added to the end of the feature vectors in the recognizer and its effect on the recognition of the speech of both genders, using gender-dependent and gender-independent systems, is evaluated.