A robust phase difference based approach to the estimation of glottal closure instants and its applications

Abstract

Estimation of Glottal Closure Instants GCIs plays a vital role in glottal synchronous processing in applications such as prosody modification speech dereverberation speech enhancement glottal source modeling artificial bandwidth extension speech synthesis and detection and diagnosis of vocal fold pathologiesThese instants may be estimated from electroglottograph EGG and speech signals Since the EGG is devoid of the vocal tract system information estimation of GCIs from these signals is easier. However an EGG signal corresponding to a given speech signal might not always be available owing to the relative difficulty in attaining it This poses a need for the estimation of GCIs from speech signals Further as evident from the afore mentioned applications the estimation of GCIs may not be restricted to speech collected in a noise free environment over a microphone The speech signal might be degraded due to the presence of spectral and temporal perturbations making the task of identifying GCIs more challenging This necessitates a GCI estimation algorithm that is not very sensitive to any degradation in speech or EGG Therefore the current work focuses on deriving a robust algorithm to estimate GCIs from clean and degraded EGG and speech signals by modifying the conditional group delay Phase Difference PD algorithm initially proposed for use on clean speech to make it suitable for degraded speech/EGG signals as well Further to test the suitability of the PD algorithm for use in a practical scenario where clean speech is rarely available two applications related to prosody modification of synthesized speech are considered. newline