A new approach for glottal closure instants detection from speech signals

Abstract

Glottal Closure Instants (GCIs) are defined as time instants of significant excitation of the vocal tract system during the production of speech. Non-stationary nature of excitation source and vocal tract system makes accurate identification of GCIs, a difficult task. All the existing methods for GCI detection require knowledge of voiced activity regions and a rough estimation of pitch frequency. In this paper, we propose a novel method based on time-frequency analysis that can be employed on entire speech signal to detect GCIs without any prior information. The proposed method automatically detects regions of glottal activity and determines an estimate of pitch frequency for each voiced activity region. The proposed method locates GCIs with high accuracy and reliability. Simulations are carried out on speech signals taken from CMU-Arctic database and the short time Fourier transform (STFT) has been used as a time-frequency analysis tool. The performance of the proposed method based on time-frequency analysis is found out to be significantly better than existing methods.