Automatic Arabic digit speech recognition and formant analysis for voicing disordered people

摘要

In this paper, analysis of speech from voice disordered people is performed from automatic speech recognition (ASR) point of view. Six different types of voicing disorder (pathological voice) are analyzed to show the difficulty of automatically recognizing their corresponding speech. As a case study, Arabic spoken digits are taken as input. The distribution of first four formants of vowel /a/ is extracted to show a significant deviation of formants from the normal speech to disordered speech. Experiment result reveals that current ASR technique is far from reliable performance in case of pathological speech, and thereby we need attention to this.