The objective of this paper is to design and set up an efficient lie detector that can detect lies in spoken sentences in an object language. Usually lie detection make use of invasive methods that extract different physiological signals such as EEG etc., from the subject’s body and analyze them to decide on the presence of lie in the utterances. This work proposes an efficient lie detection technique based on the non-invasive method that involves only recording the subject’s speech utterances. Discriminative and meaningful features are extracted from the speech and classifiers are built based on SVM to discriminate between truth and lie. The classifier will be trained efficiently so that a better performance can be obtained compared with existing lie detectors. This paper aims to exploit the psycho-neural aspects and dependence of speech signals to predict and detect the presence of lies in isolated speech utterances.