Waveform Analysis and Feature Extraction from Speech Data of Dysarthric Persons

Speech recognition systems provide a natural way of interacting with computers and serve as an alternative to the more popular but less intuitive peripherals (input / output devices). Tools employing the techniques of Automatic Speech Recognition (ASR) can be extended to serve people with speech disabilities so that they can overcome the difficulties faced in their interaction with general public. An attempt is made here to achieve this goal by mapping the distorted speech signals of people with severe levels of dysarthria to that of a normal speech and/or less severe dysarthric speech. The analysis is carried out by comparing the speech waveforms of the people with and without communication disorders and then extracting the features from the audio files. The differences in time, duration, frequency and PSD are used to facilitate the mapping of unintelligible speech data to intelligible ones. When reasonable accuracy levels are achieved in this mapping, the normal voice can be used as the substitute / surrogate of the original distorted voice.