One of the important features of speech processing is Pitch detection. We find application like speaker recognizing, vocoder systems, instructions to the hearing impaired, extracting information about health of a person and determining if a person is suffering from any kind of illness. Also, emotional state of a person can be predicted from voice, as voice in normal condition differ from voice when happy, sad, stressed etc. Various pathological disorders related to nasal, respiratory and larynx can be determined using acoustic parameters. Hence, accurate and robust pitch detection is necessary. The paper involves determining pitch of the speech signal by the pitch detection algorithm - 1) Autocorrelation method, 2) Cepstrum Method, 3) Simplified Inverse Filtering Method 4) Data Reduction Method. An attempt is made to identify and classify the voice samples into pathological and normal voice using kNN classifier. Database used here is Saarbruecken Voice Database, created by Institute of Phonetics, Germany. A comparative study of several pitch detection algorithms is made in terms of accuracy and time to evaluate the sample.
Indian Member 40.00
Others Member 3.00