A Computer-Based Analysis of Acoustic Properties of Sri Lankan Pirith Chants Using Voiced to Unvoiced Ratio and Probability Distribution Functions

Abstract

Pirith is a special type of chanting; believed to be a protective doctrine preached by the Load Buddha in Pali language. Generally, a voice signal consists of the fundamental frequency, F0 and a series of harmonic frequencies called as formants, Fn. As reported by several other studies, characteristic formant frequency distributions are identified in chanting, contrast to normal speaking. This work is dedicated to developing a phonetic picture on Pirith chants investigating probability distribution of formants and quantify voiced to unvoiced ratio utilizing computer-aided tools. In this study, 25 samples of each Rathana, Karaneeya Metta and Angulimala Suttas recited by male monk chanters were recorded using high precision microphone array and then subjected to splitting of smaller voiced segments of frame length 10ms sampling at a rate of 44.1 kHz. In the computational speech model, a pre-emphasis filter is applied to the sampled time series of voiced segment to cancel out the effect of glottis. Then frame-by-frame analysis was used with hamming windows and liner predictive coding (LPC) and auto correlation to extract the formant values. Voiced to unvoiced ratio is assessed using zero crossing rate and energy content of the acoustic signal. Results of the Voiced to Unvoiced ratio over 75% of voiced frames in all types of Suttas despite number of monks involved in chanting. Having a high percentage of voiced frames interpret strong contribution of vibrating vocal folds involved in chanting of Pirith Suttas. Further, Probability Distribution Functions (PDFs) of each Sutta is generated and compared for first five formants. Angulimala Sutta and Ratana Sutta show similar patterns in terms of PDFs while Karaneeya Metta Sutta indicates a clear discrepancy demonstrating a unique set of characteristics.