Comparative performance study of several features for voiced/ non-voiced classification

This paper presents a comparative performance study of several time domain features for voiced/non-voiced classification of speech. Five classification schemes have been developed by combining one or two features amongst: Energy (E), Zeros Crossing Rate (ZCR), Autocorrelation Function (ACF), Average Magnitude Difference Function (AMDF), Weighted ACF (WACF), and the Discrete Wavelet Transform (DWT). The development of these classifiers was based on the selection of the lowest number of time domain features which allow voicing decision without the need of any frequency transformation or pre processing approaches. The performance of the classifiers has been evaluated on speech data extracted from the TIMIT database. Two different noise types: White and babble, taken from the NOISEX92 database have been incorporated to validate the developed classification schemes in noisy environments. An overall ranking of these classifiers for high and low Signal to Noise Ratios (SNRs) have been established based on the average value of the Percentage of classification accuracy (Pc).