In a communication system, when sending an information signal from the sender to the receiver, the signal will be mixed with other unwanted signals. The unwanted signal is called noise. One form of information signal is a sound signal. Random sound signal. The pattern or characteristics of the signal will change from the original if the sound signal contains noise. The more noise it contains, the more the pattern changes. Therefore, a method is needed to reduce the noise contained in the voice signal before it is estimated. In this study, we will compare the wavelet transform and fast Fourier transform (FFT) methods to reduce noise in voice signals. From the test results, the Wavelet Transform method was obtained using Soft Threshold using Mother Wavelet Coilet 5, which gave the best performance; noise was completely eliminated; and the smallest information signal that was eliminated was 8.1606 Watt or 9.117 dBW compared to other Mother Wavelets. Meanwhile, with FFT, noise that is spread over the entire frequency range cannot be completely eliminated.