A Robust Non-Parametric and Filtering Based Approach for Glottal Closure Instant Detection

Authors : Pradeep Rengaswamy, Gurunath Reddy M, K. Sreenivasa Rao, Pallab Das Gupta

Speech is filtered using Raised Cosine Filter (RCF). The RCF is a pulse shaping low-pass filter widely used in digital communication for minimizing the inter symbol interference. Initially, the RCF filtered speech signal is thresholded to detect the voiced/unvoiced regions. In each identified voiced regions, the peaks corresponding to the GCIs are emphasised by non-linear filtering. Followed by a novel average epoch interval detection method based on the histogram and GCI detection based on peak picking approach is proposed.

Usage:

Through Matlab command prompt call the GCI detection method with any speech wave file as

GCILoc= GCIDetection(filename)

ex: GCILoc= GCIDetection(arctic_a0007_speech.wav)

Where arctic_a0007_speech.wav is the speech wave file, GCILoc is the vectore of GCI locations in seconds.

The file arctic_a0007_GCI.txt is the ground truth GCI of arctic_a0007_speech.wav

Manually annotated GCI dataset for CMU-ARCTIC dataset [1] is available in CMU folder for three different speakers folder (bdl, jmk, slt). The dataset consists of 100 utterances selected randomly for each speaker. Each utterance is provided with speech signal, EGG signal and the manual annotation.

Source code is tested in Matlab 2015A, Windows Environment.

[1] J. Kominek and A. W. Black, “The CMU Arctic Speech Databases,” in Fifth ISCA Workshop on Speech Synthesis, 2004.