Current Trends in Signal Processing Volume 2, Issue 1, April 2012, Pages 11-25 _________________________________________________________________________________________ ISSN: 2277–6176© STM Journals 2012. All Rights Reserved Page 11 Analysis of Speech Using Different Methods Prof. S. China Venkateswarlu 1 *, Dr. K. Satya Prasad 2 , Dr. A. SubbaRami Reddy 3 1 Research Scholar, 2 Rector, JNTUK, Kakinada, India 3 Dean, LBRCE, Mylavaram, India *Author for Correspondence E-mail: cvenkateswarlus@gmail.com 1. INTRODUCTION The problem of automatically separating music signals from speech signals has been extensively studied. In general, approaches to this problem consider a small set of features to be extracted from the input signals. These features are carefully chosen to emphasize signal characteristics that differ between speech and music. This project combined two well-established features used to distinguish speech and music, as well as a third more novel feature. Once the typical values of these features were defined by a set of training data, a decision system for classifying future samples was chosen. Here, a simple k-nearest neighbor algorithm was implemented to determine whether an incoming sample should be considered speech or music. The implementation considered here treats each sample as a whole and labels the entire sample as either speech or music [1–3]. 2. SIGNAL FEATURES A large number of signal features have been employed for the problem of discriminating speech and music. This paper used two well-established features, the zero-crossing rate (specifically the variance of this rate) and the percentage of low-energy periods relative to the RMS value of the signal. The third feature used was a novel measurement of the residual error signal produced by linear predictive coding. These three features were combined to improve the robustness of the classification system and to hopefully balance out any ambiguities in any single feature set. 2.1. Variance of Zero-Crossing Rate For this feature, the number of time-domain ABSTRACT In speech analysis, the voiced-unvoiced decision is usually performed in extracting the information from the speech signals. In this paper, two methods are performed to separate the voiced and unvoiced parts of the speech signals. These are zero-crossing rate (ZCR) and energy. In here, we evaluated the results by dividing the speech sample into some segments and used the zero-crossing rate and energy calculations to separate the voiced and unvoiced parts of speech. The results suggest that zero-crossing rates are low for voiced part and high for unvoiced part whereas the energy is high for voiced part and low for unvoiced part. Therefore, these methods are proved effective in separation of voiced and unvoiced speech. Keywords: Implementation, low-energy, root mean square (RMS) power, speech, variance, voiced, unvoiced, ZCR: zero-crossing rate