Approved: CERTIFICATION OF APPROVAL LIE DETECTION USING VOICE STRESS ANALYSIS

In the existingtechnology, there are severalways for lie detection. Voice Stress Analysis
is one of the methods for lie detection. It measures the amount of stress in the subject's
voice to detect whether the subject is being truthful or not. The VSA is reported to be
cheaper, easier to use, less invasive, less constrained in their operation and more accurate
than polygraph with 90 to 95% accuracy. In this project, the theory of Voice Stress
Analysis is used to detect deception. Developing the software of lie detection using
MATLAB will help in the investigations of law enforcers to detect deception. There are
two approaches that are used in this project which are frequency based system and energy
based system. The frequency based system detects lie using the variation in the output
graph. High variation indicates low stress or 'truthful' voice and low variation indicates
high stress or 'lying' voice. The energy based system however, detects deception in the
energy spectrum of the voice at 20 to 40 Hz frequency. A flat waveform indicates hard
stress while a sharp waveform indicates low stress. Using MATLAB, the voice is
processed and both methods are applied. The results for both methods are then compared
with the result of a polygraph test. It is found that both methods have different results on
certain voice samples and also similar results on certain voice samples. However, the
energy based system has more similar results than the frequency based system. Thus, it is
more accurate.