Improved Epoch Extraction from Telephonic Speech Using Flat-Top Cosine Windows

Epoch extraction from the telephonic speech signal is challenging due to the attenuation of the low frequency information. The present work focuses on the strengthening of the low frequency information in the speech signal by utilizing the properties of Hilbert envelope and filter out the vocal tract resonances by an FIR filter with the center frequency at 0Hz and is based on the flat-top cosine window. The properties of the flat-top cosine window are effectively exploited in the extraction of epoch information from the Hilbert envelope obtained from the speech signal. The proposed method shows improvement in identification rate and identification accuracy when compared with the performance of ZFF of the Hilbert of speech signal for both clean and telephonic speech signal.