Speaker Independent Speech Recognition for Diseased Patients using Wavelet

Speech is the primary mode of communication among human being. Patients of Parkinson, paralysis are unable to speak fluently. It is difficult to understand meaning of words spoken by the patient. Vocabulary of requirements of patients is created by collecting speech utterances from different persons. The experiment was performed in speaker independent method. By applying vector quantization algorithm on MFCC database of code is created which is used as training module. Speech utterance to be recognized is recorded in real time without using any ideal recording environment. MFCC of test speech are compared with training module using Euclidean distance. The signal that gives minimum average distance is considered as matched speech. Another objective of experiment was to find best suitable wavelet for speech recognition. The present experimentation gives 89.33 % accuracy for biorthogonal wavelet.