Comparison of Wavelet and Hybrid (ICA and Wavelet) Methods in Separation of Sound Frequency in Forensic Audio Models

Abstract

Audio forensic is a process to improve an authenticity of a voiceprint evidence. But we must understand that
voiceprint has a lot of noise inside it. In order to get a better quality of a voiceprint evidence, we need to reduce the amount
of noise. There are several noise reduction methods that we can use. In this thesis, we will make a program with MATLAB
by using two methods. First is wavelet transformation and then the hybrid process between Independent Component
Analysis (ICA) and wavelet transformation. Wavelet transformation use high pass filter and low pass filter to reduce noise
from a voiceprint. And ICA has a principle to estimate individual signal from mixtures of signal. After we use these
methods to reduce noise from a voiceprint, we will see which methods work better for noise reduction process. The result
from these two methods will be presented in SNR (Signal to Noise Ratio) output. From this research, wavelet
transformation SNR output is higher than hybrid method SNR output. We can conclude that wavelet transformation is
better than hybrid method for a noise reduction process. The best SNR output from this research is at level 5 with average
value of 6.7274 dB for 30 seconds sample, 6.1256 dB for 60 seconds sample, and 6.0296 dB for 90 seconds sample