Study of Pitch Detection Methods

Speech is produced by vibrations of the vocal cords and the rate of vibration of the cords is called pitch. Pitch is a feature of vocal speech which may contain speaker specific information. Pitch gets affected by various factors one of which is length of vocal folds. Various methods have been developed for pitch estimation which can be classified as time domain method, frequency domain method and hybrid method. Pitch estimate can be evaluated in time-domain by identifying periodicity features within the sound wave. The most commonly used time-domain methods include: parallel processing method, autocorrelation and Average Magnitude Difference Function (AMDF) methods. Frequency-domain pitch estimation can be evaluated by identifying certain features within the short-term spectra of speech signal. Hybrid Methods have features of both time and frequency domain. This paper describes some of the time domain and frequency domain methods for pitch estimation and its applications. In this paper autocorrelation method under time domain has been implemented using MATLAB software and the results obtained are verified using PRAAT software.