Publication

Abstract : Various types of vocal disorders result in impairment of human voice. Mostly, laryngeal disorders cause whispered speech. Rectification of the whispered sound is a challenging problem. The Fourier power spectrum reveals that whispered voice has less of sharp peaks and the fundamental frequencies are more spread out than that for normal voice. This implies that whispered voice fundamental frequencies are less specific and spread over wider range. The challenge of conversion of whispered speech to normal one is addressed by many. Existing methods suffer from one drawback or the other. In this paper, we report the outcomes of our experiments to improve this situation. A few rectification techniques were applied to the whispered speech signals. Our novel methods were to employ Ordinary Deconvolution Procedure (ODP), ‘Blind’ Deconvolution Method (BDM) and ‘Unsharp Masking’ Technique (UMT). UMT is a modified version of the erstwhile technique, used in 2-D image processing in the 1-D acoustic signals. The selective tailored enhancement of the higher frequency components, thus achieved, made the whispered signals more discernible. In the case of ODP, the main challenge is that of determining impulse response. Hence, we adopted the BDM with some encouraging results. The vowel sound submerged in the whispered voice could be discerned. The primary results looked promising enough to proceed further in these lines.